Ease of Access to An Alternative Food Source Enables Wallabies to Strip Bark in Tasmanian Pinus radiata Plantations
Round 1
Reviewer 1 Report
I'm afraid that I couldn't find the major improvement in the manuscript from the previous version. I'm sorry for my negative review again. I hope my thoughts below could be a help to improve the manuscript.
The purpose/aim of the manuscript is still unclear in Introduction. Why did the authors focus on spatial scale? What does the spatial scale specifically mean here? How and why were intra- and inter-site factors distinguished? Better explanations and reasoning is needed.
The analytical step is also still unclear. This is partly because it is poorly structured. For example, it is difficult to understand how many variables were actually used in regression model. It's better to distinguish data compilation process and the explanations of model variables. Especially, the words "Response variables" and "Explanatory variables" should be used after defining analytical framework. In addition, the model selection procedure (how to define models and how to use information criteria) is not explained and no results were shown.
In a statistical view point, the regression model includes too much explanatory variables (though I couldn't know exact number) compared with sampling size (60 and 12). I'm not convinced by the results because of this problem. I recommend to reconsider the analytical framework.
I appreciate the author's effort to make Wotherspoon 2004 publicly available (though I couldn't see it yet). However, my major concern is not solved. My point was that the authors must explain how general Wotherspoon's findings are and/or how compatible them with the current study. In this sense, the current arguments are insufficient and unconvincing.
Author Response
Reviewer 1
This reviewer had two major problems with our manuscript. The first had to do with ‘spatial scale’, the reviewer not being able to comprehend how and why intra-site and inter-site factors were distingushed. In our revision, we now try to make this clearer by splitting Table 2 into two sections, using subheadings. The plot-specific (intra-site) variables are those that were measured in each of the 5 plots at each site. Thus, the percentages of the individual components of the vegetation such as grass, bracken, moss, and individual species, as well as the percentages of bare ground, rock and woody debris, and tree characteristics such as height, no. of branches in the whorls, internode length, etc., all fall into the ‘plot specific’ category. This also includes the minimum soil temperature that was measured with a Thermochron ‘ibutton’ in the centre of each plot. The site-specific (inter-site or inter-plantation) variables include geographical features such as altitude, rainfall, maximum and minimum air temperatures, for which meaningful measurements or estimates can be obtained only in the broader scale of a plantation. Therefore, to enable modelling of the percentage of bard girdling to be carried out when temperatures differences such as Tdiff was considered as a potential explanatory variable, it was necessary to take averages over the information in the individual plots, thereby reducing the sample size from 60 to 12.
The second major problem this reviewer had with our manuscript related to the statistical analysis. The reviewer believes that we may have included too many explanatory variables in the model and that it was not clear how many of these variables were actually used. We apologize for our failure to make the methodology clearer. The multiple regression models that we employed were a combination of forward stepwise regression and ‘all possible’ regressions, options that are available in SAS PROC REG, the software that we used. Forward stepwise regression enables the modeller to build up a model from scratch, starting with the explanatory variable that contributes the most to explaining the variation in the response variable. Sometimes, the best model containing two explanatory variables will not contain the single best explanatory variable. That is the value of using ‘all possible regressions’, which examines all possible models with a single explanatory variable, then all possible models with two explanatory variables, etc. The fact that the best model at the plot level as adjudged by BIC contains only a single explanatory variable (BBM) shows that we do not have too many explanatory variables in our models. Indeed, we have no more than three explanatory variables in any of the models in Tables 3 & 4. In our revised manuscript presented here, we now make it clear that we have used a combination of forward stepwise regression and all possible regressions and we thank the reviewer for persistently criticising the lack of clarity over two versions of this manuscript.
A third issue raised by Reviewer 1 related to reference [3], an unpublished report by K. Wotherspoon. We tried desperately to convince Sustainable Timber Tasmania, the organization that superceded Forestry Tasmania, to allow this privately commissioned report to be made available on the Internet. Regrettably, they felt unable to accede to this request since Wotherspoon’s study was financed by a private company rather than from public money. However, they are allowing us to refer to and quote some of the findings from that preliminary study and we have done so: the material in the Discussion from lines 299–314 and on lines 333–334 all relate to material in Wotherspoon’s report.
Reviewer 2 Report
The study by Smith et al presents an analysis of the factors that affect damage to plantations of Pinus radiata in Tasmania. The manuscript is nicely written, and I mostly have questions regarding the description of the methods and the statistical analyses (mostly clarifications, I believe). I provide my suggestions as specific comments below.
Specific comments
L21-24: can this be phrased a bit more specifically indicating the direction of these relationships, i.e. ease of access variables increased the percentage of bark stripping and variables related to hindrance reduced it?
L29: I was missing here (end of the abstract) a bit more of a conclusion sentence, or something on the implications of the study. For example, you found that access is key to the amount of damage by wallabies, so making access more difficult could help prevent damages (or something like this).
L33: I had a bit of trouble with the word order here – but maybe I am just overthinking it :) how about: “softwood species widely planted worldwide”?
L82: was the range in damage severity an a priori criteria to select the sites? If so, what data was this based on? It would be interesting to know if there was a prior assessment of damage at your sites, especially since you come back to this in the discussion when you mention that damage can vary widely in different years. In a similar way, are there any estimates of wallaby abundance that could be related to your sites? Also, how big were the plantations? Could this be an additional factor in explaining damage (e.g. if they are widely different, you may expect differences in damage as they would provide a more heterogeneous or homogeneous landscape). Maybe this information could be included in Table 1.
L85: so plots were roughly 12x10 m in size? Might be good to clarify this here (and see also my question below)
L88: what was the average/range in distance from the main road to the plots?
L93: per plantation or per plot? Does this mean that you sampled four trees per plot (20% of 20 trees in a plot)
L96: where there any cases where bark stripping could not be unequivocally attributed to wallabies?
L98 (Table 1) Tdiff – is this an average difference?
L101: It might be helpful to show the scale of the map, and also an inset with the location of Tasmania relative to Australia :) Also, it might be helpful to show additional information in the map – for example you could make point size proportional to the damage found in the present study. Maybe instead of numbers for each site you could use the same acronyms as in Figure 2?
L107: not sure I understand this explanation of the estimates of damage. Does this mean an estimate of the total amount of bark removed for a whole tree? Or an estimate of the perimeter damaged at a certain height of the tree? It might be good to have pictures of this in an appendix, showing different degrees of damage
L107: was age estimated to age classes? If so, how many classes? How was this information then included into the analyses?
L118: how long were these transects? (the plots are only 10x12 m in size if I understood correctly). How were they placed within each plot? How were the visual estimates conducted (i.e. percent cover along the line transect or within an area defined by the transect)?
L119-122: maybe just refer to Table 2 in brackets (not directly within the sentence): “The variables (measured as percentage cover) that potentially aided access were (Table 2): ...”
L133: “...two functional groups...”
L158: you have many potential explanatory variables (for example at the plot level, there are 21 variables listed in Table 2, and you have 60 observations). Did you estimate correlations and multicollinearity (using e.g. Variance Inflation Factors) between them before building your models? This could help you reduce the set of candidate variables and build more meaningful models.
L159: why not use a mixed effect model for the plot level variables with site as a random factor?
L160: maybe refer here to the corresponding tables where the plot level (Table 2) and the site level (Table 1) variables are presented?
L166: if I understood correctly Table 2 presents only the plot-level variables
L179: remove “which”?
L200-201: this could be presented in the introduction?
L232: it might be useful to indicate the amount of damage for each site here? And also in the text
L268-271: see my comment above about correlations between explanatory variables.
L274: it would be nice to have some examples here of which browsing mammals :)
L283-292: if seasonality in damage patters is a concern, what are the implications of conducting your surveys between Oct and Jan?
Author Response
Reviewer 2
L21–24. The sentence has now been rephrased to indicate the direction of the response.
L29. A sentence has been added which mentions limiting access, concluding that it is cost prohibitive.
L33. The wording has now been changed as suggested.
L82 (and L85 and L93). Criteria for selecting the sites were based, at least in part, on the Wotherspoon (2004) report, which was an industry-funded preliminary study that cannot be made generally available. As stated in the manuscript, sites were selected to represent a range of altitudes, rainfall and damage severity. To obtain an estimate of the degree of damage at the plantation level, spatial mapping was carried out between October 2006 and January 2007. It was then when ‘every tree in every 5th row per plantation’ was assessed (see L93). However, this is a source of confusion because it was part of an initial assessment, not part of the actual data collection in the study, which was carried out on all 20 trees within the plot that, as the reviewer deduced, was ca. 12 x 10 m in size. To avoid this confusion, the sentence containing the words ‘every tree in every 5th row’ has been deleted in the revised manuscript. With respect to plantation size, this varied greatly and was the subject of a supplementary study which did not yield any positive results.
L88. The average and range of distances from the main road to the plots has now been added to the revised manuscript.
L96. No bark stripping cases due to other causes were reported.
L98. Yes, Tdiff is an average difference and the abbreviation ‘Ave.’ has now been added to the heading.
L101. A scale has now been added to the map. It was felt that an inset indicating the location of Tasmania relative to the rest of Australia is unnecessary, given the ease with which readers can obtain accurate information in this regard (e.g. Google Maps).We have adopted the acronyms for each site instead of numbers, as suggested. Making the point size proportional to the damage seems like overdoing it; however, we have added the percentage damage to the caption of Figure 2, as suggested by the reviewer (see L232).
L107. For each of the 20 trees in a plot, the proportion of the tree’s perimeter that had its bark stripped was measured. The %girdling score for that plot was the average over the 20 trees. In addition to this %girdling score, other measures of damage were made, such as damage height and area, but these were deemed to be less reliable than %girdling for subsequent modelling.
L107. Data related to age classes did not give useful results.
L118. There were three 1 m2 quadrats randomly located along a diagonal transect within each plot. This information has now been added to the manuscript.
L119–122. Refer to Table 2 in brackets. That has now been done.
L133. The term ‘functional groups’ has now been deleted.
L158. Correlations between potential explanatory variables and the use of VIF’s were not needed, as regression analyses were carried out using forward stepwise regression and ‘all possible regressions’. Forward stepwise regression enables the modeller to build up a model starting with the explanatory variable that contributes the most to explaining the variation in the response variable. Sometimes, the best model containing two explanatory variables will not contain the single best explanatory variable. That is the value of using ‘all possible regressions’, which examines all possible models with a single explanatory variable, then all possible models with two explanatory variables, etc. We have now made it clear that we have used these multiple linear regression techniques in our revised manuscript. The fact that the best model at the plot level as adjudged by BIC contains only a single explanatory variable (BBM) shows that we do not have too many explanatory variables in our models. Indeed, we have no more than three explanatory variables in any of the models in Tables 3 & 4.
L159. It is not clear to us why a ‘mixed effect model’ would be appropriate, as the data obtained in this study clearly suggest a regression model, not an analysis of variance model. The 12 sites, as it were, provide the information for replication.
L160 and L166. As a result of these comments, Table 2 has now been given subheadings which allow plot-specific and site-specific variables to be listed separately.
L179. “which” has now been removed.
L200–201. This material about the hypothesis involving sugars and starch has been moved to the Introduction as suggested.
L232. As suggested, the amount of damage (as %girdling) has been added to the text in the caption to Figure 2.
L268–271. The use of all possible regressions is probably the best way of avoiding getting misleading results from multiple regression analysis. It is necessary to use common sense and not accept models solely based on stopping rules and criteria such as r2, AIC, etc.
L274. The words “browsing mammals” are as they appeared in Wotherspoon (2004). He was referring to wallabies and possums, and we have added those words to the revised manuscript.
L283–292. The surveying between October 2006 and January 2007 enabled the quantification of the damage that occurred in the preceding winter and spring. In Australia, winter is considered to occur from 1 June to 31 August, and spring from 1 September to 30 November.
