Browsing Damage on Scots Pine: Direct and Indirect Effects of Landscape Characteristics, Moose and Deer Populations

Round 1

Reviewer 1 Report

This is a good study that enlightens the grazing processes by extant wild herbivores in European boreal latitudes. It also illustrates a good use of Path Analysis as a statistical tool.

I have one concern for the whole manuscript. The continuous use of the word "forest" to describe the dominant landscape of Scots  pine plantations in Sweden is, to me, not descriptive of the ecological quality of these stands. As the authors rightly contextualize in the MS, these are commercial timber plantations, and not a natural ecosystem (which may even not be a closed canopy forest - check https://www.researchgate.net/publication/273108489 or https://doi.org/10.1093/oso/9780198812456.001.0001). In the same sense, a sunflower crop field is not an annual pasture, even if it is fully covered with grassy annuals. In that sense, I demand a thorough revision of the MS that corrects this terminology, for it otherwise gives the overall impression that we are talking about forests here.

Other comments:

L10 Add "economically" before "valuable". The reader will otherwise believe that you refer to other types of values, notably environmental ones.

L252 Add "growth" between "vegetation" and "period".

L315 typo at citation 72

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper deals with browsing damage on Scots pine by moose and deer populations in Sweden. The paper attempts to disentangle direct and indirect browsing effects of moose and deer populations using a path analysis approach.

The topic is interesting from an applied perspective as browsing can have a strong negative effect on forestry production and woodland management.

The paper is well written, and the presentation of results is good overall.

The authors claim on the advantages of path analysis over standard regression approaches to differentiate this paper from others published on the topic. I have serious concerns that this has been achieved, which questions the results of the paper (see below).   

Line 46 & 51. It is true that regression cannot prove causality, but neither can path analysis. In a path analysis all you can do is propose a causal model, make certain assumptions and test that the data don’t seem to break the model or assumptions.

I have serious concerns on the design of the path analysis and so the core of the narrative of the paper. The path (a) is pretty weak, from what I can make out – if it really is just the set of “background” variables, then all they are doing, in effect, is testing whether the cervid population data is useful for improving predictions. And for that, a plain regression model would be just as useful. The authors could test this by running a regression analysis on the whole set of variables and discussing on the different results produced by path and regression analyses. I think this is a more robust and less subjective way to interpret the relationships between all variables.

Furthermore, some of the possible “indirect effects” seem pretty obvious, e.g. forage availability is surely only ever going to be an indirect effect…and as the main endogenous variable is “browsing damage”, you’d imagine the presence of something to do the browsing would be a necessity.

I think the reader might find the reasoning behind the design of the path analysis subjective and thus undermine its value compared to a well-conducted regression approach in which the selection of variables has been thoroughly thought through. Again, a side-by-side comparison between a path analysis and a regression approach could be very useful to disentangle the correlation between variables.

I am not convinced by splitting the dataset into northern and southern MMAs to further perform path analyses on all MMAs, northern MMAs and southern MMAs, to conclude that the patch analysis of the southern MMAs resulted in a poor fit between their hypothesised model and the data. A regression approach could easily have included a latitude variable (e.g., north versus south) to test for differences between boreal and nemoboreal vegetation types or well thought interactions between predictors.

 

All in all, I believe that the narrative that justify the path analysis design of this study is subject to interpretation and thus the results and conclusions of the paper. A more robust approach could be achieved comparing the results of a path analysis against a regression approach and discussed on the differences between them and published studies on the topic.

 

I hope this helps.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The new version of this paper has not addressed the concerns of my review in a satisfactory fashion.

On my comment about the subjective nature of their path modelling, author rely on their expertise to justify its validity. To be frank, all this causal stuff requires quite a lot of subjective assumptions which can be tested…but that still ignores that alternative subjective assumptions can also be tested and might actually be better.

My biggest concern is still there, more food causes more animals causes more browsing and is path (a) still weak, even if it is the “strongest”. The approach of “a poor fitting” of a path analysis applied to north and south zones to support the hypotheses is not convincing.  

Of course, path analysis is really just a particular sequence of regressions anyway, and I would certainly explore different regression models in this sort of analysis to try to understand what was going on. This is another kind of extension of “(multiple) regression analysis”.

Certainly, whatever the authors found in a path analysis could be tested via regression, even just as a sense check… and still it will not mean causality but will ensure that there are sufficient caveats in the paper to articulate the complexity of the relationships.

Author Response

Response to reviewer comments:

 Reviewer #2:

The new version of this paper has not addressed the concerns of my review in a satisfactory fashion.

On my comment about the subjective nature of their path modelling, author rely on their expertise to justify its validity. To be frank, all this causal stuff requires quite a lot of subjective assumptions which can be tested…but that still ignores that alternative subjective assumptions can also be tested and might actually be better.

My biggest concern is still there, more food causes more animals causes more browsing and is path (a) still weak, even if it is the “strongest”. The approach of “a poor fitting” of a path analysis applied to north and south zones to support the hypotheses is not convincing. 

Of course, path analysis is really just a particular sequence of regressions anyway, and I would certainly explore different regression models in this sort of analysis to try to understand what was going on. This is another kind of extension of “(multiple) regression analysis”.

Certainly, whatever the authors found in a path analysis could be tested via regression, even just as a sense check… and still it will not mean causality but will ensure that there are sufficient caveats in the paper to articulate the complexity of the relationships.

 

Response:

We acknowledge that reviewer 2 has an important aspect regarding the subjective nature of our model that we would like to clarify. Yes, there are several models that can be tested and we would like to clarify more on why we choose to test only this specific model.

First, the current discussion in Sweden regarding browsing damages within young forest stands centres around the question whether it is more efficient to increase forage availability than to (further) lower the moose population. The root of this discussion in Sweden lays partly in previous multiple regression based models which indicated a stronger relationship for pine (or forage) availability than moose density. This discussion was reinforced by a recent report that concludes that browsing damages have not been affected although the moose population has been reduced by about 20 % between 2012 and 2020. Thus, in the public discussion, basic causality such as “fewer moose will lead to fewer damaged trees” is questioned among moose managers on various levels.

Secondly, path analysis has not been used to test a model of browsing damages in a similar context. We wanted to learn more about how path analysis can be used to elaborate more on causal models that may guide decision makers. The multiple regression method is designed to find the best model that could explain the data, irrespective of causal links. Thus, if latitude is a better predictor than temperature then latitude may kick out temperature from the regression model even though temperature might be the actual causal link. In path analysis, it is assumed that the focus is on causal links and that the model is built upon previous empiric and theoretical understanding.

Thirdly, our article would have been lengthier and without a clear focus if we included a comparison with a multiple regression. In fact, we performed additional multiple regressions at an earlier stage of this project but we decided to exclude that from the manuscript as there are several other studies within this topic that are based on multiple regression providing similar results. We therefore refer to these studies instead of adding a similar regression model within the article. We elaborated on these studies in the manuscript (see line 53-62).

We hope that our paper can stimulate a scientific discussion and encourage other researchers to test other, perhaps more refined, models explaining browsing damages. This work is the first attempt of using path analysis with publicly available national data used in management to explain browsing damages across areas. It has both validity and importance to the managers and will add to the public debate by focusing on the basic causal links.

Round 3

Reviewer 2 Report

Once again, the new version of this paper has not addressed the concerns of my review in a satisfactory manner.

Author Response

We understand that you are not satisfied with our edits, as we decided not to include an additional regression analysis. Nevertheless, we added now a new section (4.4, L592-623) in the discussion, which takes up some of the issues that you raised in your previous reviews and our standpoint on them. We hope this will not only allow the reader to assess the validity of our approach, but also meet your approval and conciliate our opposing views on the matter.