Determining Statistically Robust Changes in Ungulate Browsing Pressure as a Basis for Adaptive Wildlife Management
Round 1
Reviewer 1 Report
The authors developed a test that can assess significant changes in browsing probability between inventory periods to provide wildlife managers a tool to aid their activities regarding the assessment of the impact of herbivores on the regeneration of trees in a forest. It could be a valuable contribution for forest management dedicated more toward biodiversity and their ecosystem functions. Their findings are remarkable e.g the sample size can be reduced by 2/3 without too much loss of info for wildlife managers. The inventory method combined with their evaluation tool could therefore reduce their work load and costs.
However, the MS really needs extensive editing both regarding English (language and grammar)and structure. I started correcting the English language and grammar in the beginning, but stopped at a certain point. It was actually hard to focus on the content because I often needed to reread the sentences. Often the authors also tend to elaborate too much on the point they want to make.
In the results section 3.1 the authors elaborate on their simulation results for every tree species. According to me, it does not benefit the message of the MS at all, since they are not emphasizing on the condition of every tree species later in this MS.
I guess the authors want to elaborate on the use of the test they developed and as far as I understood further on in their study, they focus more on one species (rowan) to perform their simulations (or I completely missed it, sorry). I wonder whether it is possible to reduce the extensive description per species to and some text and a table with the important parameters in it and add the rest in a supplement section?
I would dare to suggest the same for 3.2.
3.3 focusses on rowan as the most common rare tree species important as a food source for ungulates.
I personally find the discussion a bit chaotic and sometimes hard to read. I think the MS would benefit a lot from integrating the literature (more concise) better with their own results. Moreover some remarks I have for this section refer back to questions/ ambiguities I mention in previous sections (see below). e.g. the 1300 seedlings/ha-1 exclusion. Why ?( because no seedlings, only mature trees, no space for seedlings, ...?). If these sites have potential, but not enough seedlings, shouldn't they be included?
One of the reasons is evidently the poor English (I gave a few examples below), but also the extensive detailed comparisons with a few other studies, often made me have to go back in the text to relate again to which objective they were referring to in their own work. The MS would gain from integrating these studies more concise. Moreover, it would also help the reader to focus more on the essence of their objectives. for instance in section 4.3 where they could just focus on the necessary improvements they think the method needs to be even more practical and useful.
Although the authors refer to an extensive list of reference articles, they have omitted some relevant work that has been performed in France (ONCFS) on this topic.
Below you can find some more detailed remarks, I hope they will contribute to improve the readability and understanding of the MS.
Just a small note, I know it's a choice, but maybe the authors can explain to me. I wonder why they focus on the number of trees and not the number of plots to estimate the optimal sample size.
Abstract
l 9-11: Or there just wasn't browsing activity?
l 14-15: are you referring to plots or trees here?
keywords:
probabilities: sounds strange to have it as a stand alone keyword for this journal
Introduction
l 47: Exceed??
l 53-54: what do you mean with " ....to adapt flexible to changing conditions."
l56: when it comes to the impact of ungulates- one of the many examples I could give that needs rephrasing.
l 68-69: I really don't understand the sentence or what you want to express with this sentence.
l 74: ...and its temporal change. Of what? The mortality??
l 123: by (the department) of forestry?
l 126: many of the formerly mixed ...
L129: I find the abbreviation AMSL rather confusing
L131: the milder climate allows for ...
L123-140: I think this part does not need this elaborate explanation for this MS
l143: composes
l 144: was reduced
l 154-155: sentence not correct
l 158: I would rather write: Nevertheless there is no evidence that the populations of red deer and wild boar are regulated by predators.
l 162: population within the park...
l 164: Roe deer hunting was discontinued... ??
(here I stop with giving suggestions on the English language because it really is too much. For clarity, the part above still needs rigorous editing but I still made a couple of remarks where the sentences even didn't make sense.)
L175-176: why are these districts excluded?
Line 179: Why are regeneration densities of less than 1300 plants /ha ≥ 20 cm excluded? What is the reasoning behind this? Are there grids that do not contain any regeneration area, according to your definition? These questions should be addressed, according to me.
Line 183: How do you select the transect’s orientation and is this relevant for the analysis?
Line 184: Are these regular intervals constant over all plots? If this is the case, I would mention this distance. Why not just equidistant intervals?
L 185: reason why one stays away 5m from the edge?
Line 129: I find the abbreviation AMSL rather confusing.
Line 179: Why are regeneration densities of less than 1300 plants /ha ≥ 20 cm excluded? What is the reasoning behind this? Are there grids that do not contain any regeneration area, according to your definition? These questions should be addressed, according to me.
Line 183: How do you select the transect’s orientation and is this relevant for the analysis?
Line 184: Are these regular intervals constant over all plots? If this is the case, I would mention this distance. Why not just equidistant intervals?
Line 188: Is the examination of seedlings performed unidirectional or all over the plot. This will affect your sample points’ radius.
L199: recorded yearly?
Why only points 2 and 4? Justification might be needed here.
Figure 2: Some grid points do not fit within the systematic grid design. This demands further clarification.
L218: did you define BP somewhere? I seem to have missed this then?
Lines 218-231: Using a GLMM you take into account the fact that observations nested within plots are stochastically dependent. How about heterogeneity in browsing probability within a single plot. If I understand it correctly you do not allow for browsing probabilities to vary among sample points within a plot, not by using fine-scale covariates, nor by random effect structures.
Line 221: You only consider a strictly linear effect of year t on the browsing response. Unless a clear ecological rationale can be given for this decision, I would compare this model with models that allow non-linear relationships between year and browsing response (e.g. a time spline).
Line 226: random effect or random intercept there has been no mention of a random effect so far. I would advise you to provide a mathematical model formulation of your GLMM to avoid any confusion.
Line 230: and divided this BP? into its two sections? Or what did you divide? the areas?
Equations 1 & 2: second H0 ? H1
Line 244: we "subset"??
Line 258: can you just assume this?
Line 271: most current ? latest or most recent
Section 2.5.3: I'm not an expert on simulations, but I found this section hard to understand the way it is written at the moment. I suggest the authors rewrite this section, so it becomes easier to understand it immediately.
L283: isn't it the R environment?
Figure 4, panel B ? log changes for the 3-year interval period are potentially larger than those for 1-year interval periods, is this taken into account in your statistical analysis? In addition, it might be interesting to consider the overall log change (i.e. 2008-2018) as a summary statistic.
All figures displaying confidence intervals: confidence intervals are displayed, but it is not mentioned at which confidence level the intervals were obtained. I assume you used the 0.95 confidence level, but this should be mentioned.
Lines 455-485: the minimum sample size to accurately detect PP changes were calculated based on the 2015-2018 changes. Why are the 1-year intervals not considered here? This is currently not discussed, but could be interested if a forest manager desires to detect yearly changes in BP. It is mentioned in the methods section that you opt for 2015-2018 changes as it represents the most recent situation.
Results
line 301-303: the way it is written now it is difficult to understand what you exactly mean with this.
On the results on the browsing probabilities time series: What is the purpose of giving these detailed results for the different tree species within the core of the MS, if you don't use them further in your discussion? If these results are really important for the simulation results, I would give some textual info and a table and add the detailed info as a supplementary file.
L420: browsing...browsing...this is confusing
L420-421: a tree species of special interest. Redundant because you elaborate on the reason this species is interesting in the following sentence.
L436: and remains relatively constant at a level above 200..what?
I could be entirely wrong but since an improved methodology (and the authors seem to have an idea on those improvements) combined with their evaluation tool could lower the efforts of foresters to assess the impact of herbivores, why not bring the MS from that focal point? Bring the main focus in your study on the relevant species for a forest that needs biodiversity conservation (and not a commercial one where eg Spruce is important but not relevant for conservation); maybe use rowan as your indicator species? It would be a different angle, but more directed toward a certain objective.
Author Response
Please see the attached PDF.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors evaluated browsing probability and performed a detailed analysis of how many trees should be sampled to detect a statistically significant trend in browsing probability among years. While I appreciate the amount of work presented in this manuscript, I have major concerns about the survey methods, the statistical analyses and the usefulness of this work for wildlife and forest management.
First, statistical significance does not equal biological, ecological or forestry-related significance. If you tell a manager how they can accurately detect a 0.6 increase in browsing, this will tell them nothing about the change of impact on regeneration. It doesn’t tell me anything, and this is my field of study too. The authors should avoid discussing statistical significance without a proper discussion of effect size and biological relevance.They should also avoid to put so much emphasis on finding a statistically significant difference and rather seek to determine what degree of change deserves interest.
Second, the challenge of assessing browsing impacts or intensity is not to sample X number of trees. It is to cover X number of plots, over a certain area, to adequately cover the spatial variation in browsing and provide a reliable estimate over a certain area. Individual tree browsing level is not necessarily a good indicator of the actual browsing risk, because deer can highly browse one tree and exclude the neighbouring one. Several browsing assessment methods provide plot-level estimates, either directly or by aggregating tree-level measurements.
This is problematic for the application of the described method. I’ll illustrate the issue using figure 7: yes, fig. 7 tells someone how many trees to sample to find a statistically significant change. But a manager doesn’t know how many plots they will need to visit to reach that sample size in tree number, because sampling is limited to a maximum of 15 trees per plot. The section 4.3 of the manuscript tries to tackle on this issue of sampling design, but is very vague. As a side note on this, a confidence interval for a browsing intensity can be calculated, when calculating a mean over a certain plot number.
This is connected to another major issue I have with the manuscript, and it is the regeneration inventory. Contrary to the authors’ belief, large scale inventory of forest regeneration are performed all over North America. One great example of this is the USDA inventory, which includes regeneration inventories and browsing assessment (https://www.fs.usda.gov/treesearch/pubs/48367). These regeneration inventories usually lead to a stocking number, that is the number of stem per area. The method described in the manuscript cannot provide stocking, because the number of stem inventorized was limited to 15, with a variable plot size radius. This could have generated bias, for example, if some species present a clustered distribution because of seed propagation. As such, I don’t think the authors can even state that Norway spruce is the most common species in the regeneration layer. It is the most common in their sample, yes, but the sampling design is biased for the regeneration assessment. Sampling design is also a biased evaluation of browsing pressure, because it excludes plot with low stocking, which could have been caused by extremely high browsing.
I want to end this review on a positive note. The data collected here could be used to explore other questions that would not be affected by the sampling bias described above. For example, spatial distribution of browsing or the effects of neighbouring trees on browsing occurrence. It could also provide an avenue for determining how many grid should be sampled, with a proper consideration of the sampling biases described above.
Author Response
Please see the attached PDF.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
the manuscript improved a lot in structure and English language and my remarks were taken into account or alternative choices were explained adequately.
Except for a few very minor remarks, I think this MS can be published in "forests". Kind regards
line 279: remove 'the' and write "selected rowan"
line 362: I always learned that you should not start a sentence with a figure, such as "65". Maybe you can restructure this sentence
line 375 and 414: you still use "special". Is this on purpose or do you mean "specific"?
line 401: seem should be seems according to me
line 405: maybe better change into: Furthermore, rowan is considered a pioneer species....
Author Response
Please see the attached PDF.
Author Response File: Author Response.pdf
Reviewer 2 Report
I appreciate the time taken by the authors to revise their manuscript and respond to my comments. However, it does not change my assessment of their work.
My first comment regarded the inappropriate emphasis put on finding statistical significance. Aside from small changes to the manuscript, there is still a strong emphasis on finding the number of trees to detect a statistically significant change (wrongly call ‘significant change’ in the manuscript) in browsing pressure. I’m uncomfortable with this goal. Statistical significance shouldn’t be the objective. A more appropriate objective could rather be to find a X% difference in browsing pressure with a certain degree of confidence. Even if I considered the goal to be appropriate, the results (section 3.2) are entirely expected. If you reduce the sample size, it becomes statistically insignificant. And if the variation in browsing pressure is large, you need a smaller sample size. These are basic statistical observations.
My second important concern was that the entire manuscript is centred on tree-level metrics, with the justification that temporally clustered plots cannot be analyzed at the plot-level. I’m very dubious of this justification, and the authors’ response was difficult to read. I don’t want to pass judgment on an analysis I don’t fully understand. I also don’t fully understand the analyses as described in the manuscript, and I’m unsure they are appropriate. For example, lines 249-250 : “However, since the regression parameters are subtracted as logits, converting the BP change into a response (percentage) value afterwards is not possible.”. To my knowledge, retrotransformation of parameters with a logit link is possible...?
This could be validated, but again, I don’t understand how tree-level metrics can lead to improvement in inventory. I don’t see how obtaining a number of trees to sample would be helpful to managers when they want to obtain a large-scale estimate of browsing pressure. Appropriate spatial coverage of the area to sample seems to me more important when considering browsing. The authors raise the point sufficient sampling of rare species. From a management point, I don’t think we should focus on rare species to evaluate browsing pressure. Rare species could be rare because of browsing… or for other multiple reasons. They are not reliable indicators of browsing pressure.
I feel there are a context or information I’m missing. Perhaps the question/goal of this paper is relevant for the study area or Europe. To me, it currently doesn’t make sense. Either the methods are flawed as I discussed above, or the justification for this study is not clear enough.
Author Response
Please see the attached PDF.
Author Response File: Author Response.pdf
