Integrating the Rate of Moisture Loss into Needle Retention Testing to Improve the Selection of Balsam Fir (Abies balsamea) for Use as Christmas Trees

Round 1

Reviewer 1 Report (Previous Reviewer 2)

I think authors did a good job in improving the manuscript, it is much easier to read, and they answered all my questions. Therefore I have no more questions.

Author Response

Please see attachment. 

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

The data presented are valuable and should be published.  However, the statistical model presents a linear moisture loss over time.  Figure 7 demonstrates that the moisture loss over time is not linear.  Therefore, everything that depends on linearity (e.g., Tables 8 and 9) is flawed.

The good news is that an alternative statistical analysis works very well.  I have checked the following non-linear model, and the fit to the moisture loss curves is excellent:  Y = e^(-a(x-b)) + cx + d, where e is the exponential function and x is the week of the experiment.  This equation incorporates both a linear effect and an exponential component to the moisture loss.  The (x-b) component shifts the exponential component along the x-axis to where it best complements the linear component to provide the best fit to the data.  The exponential function makes physical sense because as a sample dries, the rate of additional drying should decrease. 

A complete statistical analysis can be conducted within Excel by using the Solver function to optimize fit of these parameters to the original data.  Do NOT first average the samples, but use all the data when doing the regression fitting.  Use "week" as the x-variable and branch weight (%) as the y-variable.  Create a column that calculates the modeled branch weight based upon a set of parameters, then calculate the squared deviations in another column based on the difference between observed and model-generated values.  Sum all the squared deviations to be displayed in one cell.  Use the Solver function in Excel to minimize this value while varying the parameter values (a, b, c, and d in the equation above).  Then, use the sums of squares from the non-linear curve fitting in Excel to construct an AOV for your data.  The process involves analyzing groups lumped together (e.g., Select vs. Population for NB2020 data), then use the Solver function to fit the curve to the Select and Populations groups separately.  The reduction of the Residual Sums of Squares when splitting out these two groups, vs. lumping them together, is the SS for the difference between the shape of the best fit non-linear regressions for these two groups.  Thus, you can partition the sources of error and build the AOV table using this trick.  This is a method to make statistical comparisons, but unfortunately the coefficients for the elements in the model are not as intuitively useful as the slope values.  Instead of one parameter being related to the "slope", now there are at least two (a and c) related to "slope," which is not constant, anyway.

Tables 8 and 9 are t-tests of the slope coefficients for moisture loss.  The linear model presented is statistically identical to simply taking the percent loss at a particular endpoint, such as 6 weeks.  The reason is that the samples are defined as having 100% of their original weight at week zero, and two points define a line.  If the authors wish to not do the non-linear curve fitting, then using the weight loss at a fixed endpoint is a valid and simple measure that can be used for statistical comparisons, including for AOV.  Using the linear regression slope is not valid, because it suggests linearity in the data, whereas comparisons of endpoint data do not.

Here are some very minor changes:

Line 96, change Hately to Hatley

Line 501, add an apostrophe to read branches'

In Table 3, does "most" indicate more than 50%?  If so, then say "Branches that retained > 50% of their ..."

Lines 254 - 256.  Power transformation followed by log transformation is not the right way to end up with normally distributed residuals.  If you do non-linear regression, as I have outlined above, you will find that the errors are normally distributed.  The minimization of sums of squares for a regression that fits data well (meaning that there is no systematic pattern of residuals, as your data currently have with the linear model) automatically leads to randomly distributed residuals.

Line 261 and globally.  It is a t-test, not a T-test.  You will find that statistical comparisons with the methods I have outlined will provide an approach in which t-tests become unnecessary.  You can partition out and determine the statistical significance for any of the comparisons you are interested in.   Series of t-tests are inferior to AOV for inferential purposes when you are comparing multiple groups.  An AOV for your data could simultaneously include Province, Year and Harvest.

Table 4 can be reconfigured.  Put "Final score of 'acceptable' if present" should be at the bottom of the list for the WSTC test site.  Once you recognize that most of the categories are the same at both sites, including this table seems unnecessary.  The text can simply enumerate the criteria used at both sites an mention that the WSTC test site also added the "Final score of 'acceptable' if present as an additional criterion.

This paper may also have a conceptual flaw.  It seems upon reading that the "select" groups were chosen after having collected data on their superiority.  The authors then compare this "select" group with the "population" group.  It is unclear whether the graphs and data for the "population" is composed of all the trees, including those that were "select," or not.  If the "select" group was removed from the "population" then the graphs and tables are demonstrating the phenomenon of "cherry picking".  From any data set, if you rank those that have the highest score, and then segregate them from the rest of the population, then any measures comparing these two groups will display apparent differences, even if "differences" only exist by chance.  The manuscript needs to clarify whether the "population" data include or exclude the "select" trees.

Lines 486 to 488 state that the needle retention was better for the late-harvested branches.  Were there differences in moisture loss for these two groups?  Is that pattern consistent with needle loss differences related to moisture loss for a particular sample date?  Does a certain percent of needle weight loss translate to the same probability of needle loss independent of when the branch is collected?

A further description of the phenomenon of mummification is warranted.  What was the moisture loss curve for those particular trees? 

Line 577.  Could you let us know what the other main objective is?

Author Response

Please see the attachment. 

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Postharvest needle retention is an important trait for high quality Christmas trees. It is also important to study the trait across years, as done thoroughly, because results varies from season to season. The study is based on a fair number of branches and a well-established method, but references are sparse about detached branches versus whole tree performance. This could be improved. Furthermore, there seems to be a lack of references to earlier studies of Balsam fir postharvest needle retention, and perhaps some experiences with breeding efforts for postharvest needle retention in other Abies/Picea species could shortly be included.

My most serious concern, and why I can not recommend publication of the manuscript in the present form is to-fold:

Firstly: As I understand, the experiment is built on three sites (of unknown origin/provenance or just local material), and thereby the genetic pools and site effects are confounded meaning one can not distinguish the effect of genetics and site – despite how many tests and years used. I think this is a very important fact that have to be addressed. How can one be sure the selected trees from one group are better than the others? Why select 5 superior trees from one group compared to the other two groups of material considering the uncertainty of the geographic/genetic effect? Any temperature/weather data from the sites that can support findings / tree responses.

Second: Recommending grafting and use of only seven trees in a deployment seed orchard (as I understand) is questionable. Even assumed unrelated and that pollen would be evenly produced by the clones  1/7=14% of the pollen in the orchard pool will originate from a clone itself – potentially causing severe inbreeding in seed set and tree performance. This fact has at least to be discussed.

Looking at the data, I was tempted to suggest to reframe the study and focus on year to year variation in tree/site response patterns – this is in my opinion an under-documented area, and where the study’s most powerful data-based documentation is present. But please look into the literature and get a second opinion on this idea.

Reviewer 2 Report

General Comments:

The manuscript presents a three-year study investigating post-harvest needle retention traits on balsam fir in three provinces in Canada. Post-harvest needle retention is a crucial trait for Christmas tree production, making this study highly valuable for Christmas tree breeding. However, there are some areas that need improvement, particularly in the introduction, methods section, and results presentation. The manuscript is well-written but could benefit from clearer experiment design descriptions and a more logical organization of the results section.

Specific Comments:

Introduction:

The introduction provides a good overview of the importance of post-harvest needle retention in Christmas tree production. However, it lacks a thorough literature review that highlights the advantages of the proposed methods compared to previous approaches used in related studies. Including such a review would enhance the manuscript by demonstrating the novelty and potential improvements brought by the proposed methods.

Methods:

The methods section requires more detailed descriptions of the experiment design and materials used. Additionally, it would be helpful to include information on the selection criteria used for the 63 trees across the three sites and justify why this sample size was considered adequate despite its limitations.

Results:

The results section should be reorganized to improve the logical flow, see comments in the pdf.

Specific comments see attached pdf.

Comments for author File: Comments.pdf