Representing Zooplankters: An Example from the Foraminifera
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsPlease, see comments on attached file
Comments for author File: 
Comments.pdf
Author Response
Response to reviewer 1
‘noted’ implies that the item will be corrected should the ms be accepted for publication.
About Figure 1: I consider that Figure 1 usefully integrates much of the analysis and allows easy comparison of results and avoids moving between separate figures. Legibility is adequate for modern browsers. I can provide higher dpi if requested by editor.
Line 56: noted.
Line 53: see ‘about Figure 1’.
Line 49-50: noted.
Line 42: The statement reads “…75% agreement amongst at least 4 experts…”. That is how the authors designed their study – in fact, the low minimum of 4 supports my argument about variation in identification among experts.
Lines 107-112: I think that it is within the author’s discretion to select functions appropriate to the data and intention of the analysis. The links provided in the ms fully document the algorithms.
Line 92: noted.
Line 78: noted.
Line 74: ‘adult’. Criteria that relate reproductive maturity to morphology in planktonic foraminifera have not been established. The >149 µm size fraction is an approximate estimate of the adult stage in ontogeny.
Line 67: Not understood.
Line 66: noted.
Line 113: noted
Between Lines 98-99: see response to ‘About Figure 1’.
Line 157: noted.
Line 151: noted.
About Discussion: How well can broken shells be recognized? That is answered by their position in the plots. Broken shells meet the size criterion and only those from Lomita Quarry are used. That locality is a coquina deposit (see Scott et al., 2015 in references) and shells of T. crassaformis are allocthonous.
Line 161: noted.
Before Supplementary Materials: There is no Conclusion. Geosciences author guide does not insist on a Conclusion. The final paragraph of Discussion sums up the study. Should the editor require a Conclusion that paragraph could be captioned as such.
Line 176-177: I think the statement simply offers a possible explanation of shape differences between populations. Isolation is a component in genetic divergence of populations.
Line 247: noted.
Line 246: noted.
Line 226: noted.
Line 194: Other research – not in the context of the problem of personal perception of taxa. Useful for fossil species – Yes, the example includes fossil populations.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsA manuscript is original, well organised and written, and provides an advancement in applying of machine learning (ML) methods that enable the recognition of specimens programmatically, what obviates time-consuming manual collection of specimen abundances and work of experts.
Methods are correctly explained; illustrations are of a good quality. Results and Discussion chapters represent rather Results and Interpretation. Conclusions chapter is lacking, though.
Author Response
Response to reviewer 2
No Conclusions header: Geosciences author guide does not insist on a Conclusion. The final paragraph of Discussion sums up the study. Should the editor require a Conclusion that paragraph could be captioned as such.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsIn the manuscript "Representing zooplankters: An example from the Foraminifera", Scott describes an alternate approach for defining the boundaries of a species concept using morphospace occupation rather than expert identification, with application to building training sets for automated taxonomic identification using machine learning. The idea is worth pursuing, and the manuscript presents an interesting test case using the planktonic foraminifer Truncorotalia crassaformis. However, there are two major issues that I believe need to be addressed before publication, namely:
(1) The author argues that the morphospace approach is superior to the expert approach as the latter is subjective and based on a limited visual assessment. While I agree with this in principle, in this context it is a bit of circular argument, as the specimens used to build the morphospace were presumably themselves originally visually identified as T. crassaformis by experts. Practically, any study using such an approach would need to first sample individuals using the traditional expert approach, in which case the problem that this approach is meant to circumvent has already occurred. I don't think this fact necessarily invalidates the approach described here, but I do think that the author needs to acknowledge and address it in the text.
(2) Unsupervised clustering is a rather important part of the results, but none of the methods involved in doing this are described in the methods section. The only mention is of the FKM algorithm in the caption of figure 1, which is not sufficient. As it stands, the study as described is not reproducible.
Additional, relatively minor comments follow:
- In the introduction, it should be mentioned that the strategy adopted by Hsiang et al. (whereby labels are kept only when multiple experts agree on them) is rarely practiced in the creation of training sets for automated taxonomic identification; most studies use identifications that are provided by only a single expert (or, if multiple are involved, each image is only identified by a single person). I believe this strengthens the author's argument about why the morphospace approach may be useful.
- Figure 1 - (A) Missing x,y-axis labels, legend labels, and scale bar label; (B) Missing legend label; (C) "P" of "PC2" is missing; (D-E) x,y axis labels are very hard to see
- Line 49-50, line 66, caption of figure 1, line 92 - Italicize "Truncorotalia crassaformis"
- Line 70 - Please describe the reasons why identification of T. crassaformis is currently problematic. This will make it easier for the reader to interpret the results (e.g., the statement about similarity to T. truncatulinoides in partition 2).
- Line 113 - Reformat "(Fig. 1(F,H)" to "(Fig. 1F-H)" (as in line 94)
- Line 137 - This is a sentence fragment. Same for line 151 and line 161. They should be rewritten to complete sentences if they are to stay in the manuscript text.
- Figure 2 - 3D scatter plots are very difficult to parse in 2D, with the chosen viewing angle exercising undue influence on the perception of the data distribution. I would suggest splitting this plot into several 2D plots or adding wall projections to aid interpretation. Also: Italicize "Globigerina crassaformis". "Procrustes" should be capitalized (line 140).
- Figure 3 - Same comment as for figure 2. The label of specimen #80 seems to have visual artifacts which make it difficult to read.
- Line 164 - Capitalize "Procrustes"
Author Response
Response to reviewer 3
‘noted’ implies that the item will be corrected should the ms be accepted for publication.
Item 1. I agree that personal perception is used to assemble the specimen set. This is a fundamental neurovision function essential for our interaction with objects in the environment. But personal perception is just that – private to each taxonomist. There is no circularity in advancing methods that examine the population sampled by personal perception to provide objective data for selection of training sets.
Item 2. I reject the contention that the study is not reproducible. The outline data are provided in an appendix and code to read them into the R environment is provided. Links to R packages which specify the algorithms with information on their use are provided.
Figure 1: noted.
Line 49-50: noted.
Line 70: Reference [70] explains why.
Line 113: noted.
Line 137: noted.
-Figure 2: Separate plots are not required. Target specimens in plot are coloured and it is obvious that all lie near the centroid.
-Figure 3: As for –Figure 2.
Line 164: noted.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsI accept the author's responses to my previous raised comments except for their statement that the methods are adequately described. The author is correct that all relevant citations are present, but in my opinion, all methods and algorithms should be explicitly stated in the text to prevent confusion and promote clarity. The aforementioned issue with unsupervised clustering, for instance, can be fixed simply by adding the name of the package and algorithm used (fclust, fuzzy clustering, FKM etc.) to the main text rather than only stating it in the caption of Figure 1. While the author is technically correct that the relevant information can be found, there is no reason that the reader should only encounter this information in a figure caption when it can be easily added to the text.
Minor comments:
Line 56 - There appears to be a large unnecessary space in front of "ForCenS"
Line 78 - Remove the semi-colon after "fraction" and add a period after the closing parenthesis. Remove the extra space before "They"
Line 210 - The file actually provided in the supplement is named "Scott geosciences zooplankters.RData", but it is stated here to be "Scott zooplankters outline data.Rda"
Line 214 - Unnecessary space in front of "Funding"
Author Response
Response to reviewer3 round2
Location of references to stats packages: FKM reference kept in Fig. 2 caption, which is certainly an appropriate location, and added to Methods.
