Automated Gold Grain Counting. Part 2: What a Gold Grain Size and Shape Can Tell!
Round 1
Reviewer 1 Report
some correction are marked on the manuscript.
Paper is very well written. But it shows only one possible way of technical analyses.
Page 3 – the upper paragraph.
For me it is strange sentence. I propose to cancel it. However, sensitive information relevant to exploration cannot be disclosed. Way you are writing the paper ?.
Major revision of conclusions is required.
Fig. no 8. The classification is wrong. The distinction should be made on the basis of the same factors. Not like intergranular or crystalline
Comments for author File: 
Comments.pdf
Author Response
Gentleman
Thanks for your reviewing, and I will address the issues to the best of my capabilities. Concerning the english language style and edition, I had the manuscript reviewed by the journal editing services. So I will check with the journal what can be further done by such service to improve the manuscript. Not being english-speaking myself, rewritting by myself may worsen the manuscript. I'll do my best efforts.
Cordially
Author Response File: 
Author Response.docx
Reviewer 2 Report
In their article "Automated gold grain counting — Part 2: What a gold grain’s size and shape can show", Girard et al. present novel and comprehensive findings on morphology and grain size. For this study, the authors have collated an atlas of an impressive number of acquired images from gold grains, accumulated from multiple glacial sediment surveys on the Canadian Shield. This very rich data base provides the authors with sufficient empirical evidence to obtain valuable insights and to draw well founded and partly surprising conclusions.
The interpretation of p-values obtained by t-tests needs some revision (see below), but this does not affect the excellent overall impression of this highly interesting empirical study. I therefore fully endorse this article with minor revisions (see comments below) for publication in "minerals".
Comments:
line 30: opening square brackett missing
line 31: "exclusively" instead of "only"
line 35: typo: "through" instead of "though"
line 65: please clarify which "characteristics" you mean exactly
line 145: "About 5% of the samples"
Figure 2: in the legend, fix typo: "Lodgment till (n=1356)"
line 187: "width (measured orthogonal to the maximum length)," Is the width the maximum length perpendicular, or the feret diameter orthogonal to the maximum lenght?
line 188: I guess "surface area" and the aforementioned "area" is the same? Or is the area measured in pixels, while surface area is in units of µm2?
line 222: typo in "grain size distribution"
line 223: after "log-normal": references apparently planned, but missing
line 243: In my understanding a Student's t-test is designed to verify significant differences. So "failing a Student test" might be somewhat misleading. Maybe it is clearer to write "according to Student t-tests", instead.
lines 274-276: "differences ... were therefore not very significant". Did you test this with t-tests? Then write "were not significant". If you don't base this statement on t-tests, then use another word than 'significant'. For example write "not very pronounced", or "show no distinct differences".
line 438: Please consider citing ImageJ (Schneider et al. 2012, doi: 10.1038/nmeth.2089) and PARTISAN (Dürig et al. 2018, doi: 10.4401/ag-7865) to mention two other (freeware) programs designed to compute shape parameters from particle images.
line 449: "significantly lower" - did you verify this by t-tests? If not, it is better to write "considerably lower".
line 475-486: This part needs revision to avoid falling into a mathematical pitfall. It is mainly a matter of rephrasing. (I do agree with your conclusions, but not with how you interprete the p-values.)
Student t-tests are designed to test if the means of two samples are significantly different. It is therefore an excellent tool to investigate if two samples are different. I might have misunderstood what you meant, but to me it sounds like you refer in this paragraph to the resulting p-values of Student's t-tests, when you talk about "likelihoods". Note that a p-value is not the same as the "likelihood that (the samples) represent similar populations".
The “p-value” expresses the probability that test results under the assumption that H0 is true are at least as extreme as the observed results.
Importantly, a high p-value does not prove that your samples are equal. High p-values simply indicate that your evidence is not strong enough to suggest differences. In order to prove similarities (statistical equivalence) of two samples, you would need another test with another hypothesis: the "equivalence test". See, e.g., described in Dürig et al. (2020): DendroScan – an open source tool to conduct comparative statistical tests and dendrogrammatic analyses on particle morphometry. Sci. Rep. 10:21682. doi: 10.1038/s41598-020-78698-0. Please also note that applying such equivalence tests would require you to first define a range within which you consider samples to be equivalent. (Basically, it requires calibration.)
However, for your line of argument, I don't think that equivalence tests are absolutely necessary. Just make clear that you use Student's t-tests for testing the data sets for differences. Then you can describe that you found surprisingly few differences among them, given the diversity of surveys.
For example, you could write:
"Of the 120 pairs, only XX pairs were significantly different with a level of significance alpha of 5%, and only 48 pairs show highly significant differences with an alpha of 0.1%. Considering how diverse..."
where XX is the number of pairs with p-values of 5% and smaller. (A 5% level of significance is a statistical standard approach, so it would be good to mention that number.)
In line 483/484, you write the grain size distribution "is perfectly identical", and refer to a high p-value. It is mathematically more correct to write that the distributions "show no/none significant differences". This is certainly supported by your t-test results, and remains an impressive finding, given the fact that the sites and environments of these samples are completely different.
Alternatively, you could write that the two grain size distribution-curves overlap almost completely (which I suspect is the case, if you plot them) and are therefore virtually identical. Just avoid using the high p-value as argument for equivalence. :-)
line 512: "Student t-test" or "Student's t-test" (instead of "Student test").
line 540-546: That's a surprising and very interesting finding!
line 883: typo: "than rounded"
Author Response
Gentleman
I would like to thanks you for your revision and I will take great care to do the corrections accordingly. I'm obviously not a mathematician, and the point you indicated about p-values and student test are much welcome and will be addressed seriously.
Cordially
Author Response File: Author Response.docx
Reviewer 3 Report
Your approach to automated collection of data on gold grains sounds very interesting - as I understand from you paper the particulates are not mounted and sectioned - does this present challenges with EDS given the variation in surface and hence take of angle for X-rays? I am also curious as to whether you considered X-ray tomography as an alternative method for characterisation, particularly for gold grains, as this has the advantages of a larger sampling volume and eliminates stereological bias (particularly for some of the gold grains with quite delicate/tortuous shapes) - you may have covered this already in your part 1 paper.
Author Response
Gentleman/Madame
I would like to thanks you for your review of the manuscript, corrections will be made dilligently.
Regarding the issue of poor EDS spectrums acquired from unpolished surface, this issue has been adressed in Part-1, and will be discussed in details in part-3 that will deal exclusively on chemical signature. Still, I will add a few explanation in the present manuscript.
Regarding X-ray tomography, we did a fair amount of testing and thinking on this approach, but the method has failed in regard of routine testing due to some practical limitations. These issues were partly addressed in reference #14 (Boivin et al) currently under revision and which study I sponsored. Still, I will add a paragraph on the method and results of testing.
Cordially
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Classification shown in tab. 8 should be organize according to the same distingushing features (parameters)
