Spatial Habitat Structure Assembles Willow-Dependent Communities across the Primary Successional Watersheds of Mount St. Helens, USA

Round 1

Reviewer 1 Report

line 62 - this might be true in the systems cited by nutrient availability is often very limiting early successional systems, (notwithstanding some types of volcanic eruptions)

 

121 - species richness of what? arthropod or microbes mentioned earlier?

 

line 54- 66 - I think the spread of willow on MSH is not as simple as proximity to water as captured by distance from streams, as you have described here. Willows tend to establish in drier areas away from streams in years where rainfall is abundant and cooler temperatures prevail at the time of seed dispersal and germination. Once these plants reach a certain size they develop extensive root systems that help buffer against drought. Also perched water tables at lower elevations allow for the establishment of willow thickets at considerable distance from streams (which you do not appear to sample for this study based on Fig 1). Lastly, we know from weevil exclusion experiments on upland willows that water is less a factor affecting willow size than is weevil herbivory. Sprayed plants tend to become quite large, approaching the size of riparian plants. I don't think this complexity undercuts your study, but proximity to streams could be due to them being a source for willow propagules and not just water availability.

 

119-135 - this paragraph could be organized and clarified (especially with regard to hypothesis 1) to better link to the results and discussion (specifically the competing sets of models you present below).

 

spread out of streams is driven entirely by water availability and that distance from stream essentially captures that. Willow seeds tend to establish best in wet years where germination and establishment occurs 

 

 

2.2 Do you control for the size of willow plants when estimating the species richness on each? If not, can you justify why this doesn't matter?

 

2.5 The methods described in this section were unclear to me. First, what is the response here? I presume you  are using the GPS locations of the 348 tagged plants in Fig1C in your circuitscape analysis? If that is the case, based on your map, the hap-hazardous location of these willows does not seem adequate for this kid of analysis. Wouldn't you want to sample willows in perpendicular transects from these streams at randomly chosen intervals between their source and outflow into Spirit Lake? Looking at the Fig more closely, there is a lot of spatial heterogeneity in the location of willow with respect to streams, with much more extensive thickets further away from streams as you progress towards Spirit Lake. However, the nature of your sampling does not appear to capture any of this variation at all, being restricted to higher elevation headwater areas. Second, how did you pick select between explanatory variables in this connectivity model? You say you tested combinations of stream location with slope, aspect, etc and selected stream location because "landscape resistance changed little compared to using stream locations." Does this mean a model with slope/aspect perform similarly to a model with stream location and if both perform well why did you select stream location as the "best" model other than that it supports your apriori hypothesis of water availability being the most important factor governing the spread of willow?

 

2.6 Again, the methods described in this section should be made clearer to aid in their assessment. The circuitscape model is providing a predictor to be then used in these richness models? Why did you choose which variables to include in each suite of predictors and what interactions did you choose to test? How do these three suites conceptually relate/differ to and from one another with respect to richness? Are you pitting these suites of variables against one with three models  or are you sequentially explaining variation in richness as you progress through these suites? What kind of GLM are you using (the link function, how do you combine predictors and their interactions, tests of model assumption, especially with regards to spatial patterning in the residuals)? You should clarify this and link these suites and your choices back to the hypotheses at the end of the introduction to establish a clear chain of reasoning here. 

 

Two more specific questions regarding these two models - your analysis is entirely fixed effects here, but there is likely variation in these predictor/response relationships, as well as some non-independence, between the streams. You can see this in Table 1, where stream effects differ in their intercepts. I think you would better off here using a random intercept and/or slopes for stream identity - basically a random coefficients ANCOVA model for the first model suite. 

 

239: "Prior to 239 analysis, some entries were removed on which we observed no dependent organisms." I have no idea what this means. 

 

Your use of AIC tables to winnow down combinations of large numbers of predictors (depending on how many interactions you have) can in itself lead to overfitting and selecting models with spurious correlations (this is described well in Statistical Rethinking by Richard McElreath). This is a tricky area and there is not one clearcut solution here, but it would be good if you discussed how you dealt with these overfitting risks with your method of choice. An alternative approach could be to  use random forest to first pick a subset of variables based on their variable importance scores before fitting your richness models. 

 

Line 252: "To determine GLMs and PERMANOVAs with the greatest explanatory power, we performed Akaike’s Information Criterion (AIC) model selection for all parameter suites." AIC is not telling you explanatory power, it is proxy for predictive power. Explanatory power would come from something like R^2 or RMSE. Also, you are pitting these models against one another and not against a null model in the AIC table, meaning your top model might not explain much variation in the response. A better approach would be to pit each variable (and select combinations) against a null model with AIC, and keep ones that perform better than the null. You could then do some multi-model inference based on the set of kept models and or report the R^2/RMSE of the top model or set of models as a measure of explanatory power.

 

Line 273 and throughout: You refer me to a supplement that isn't provided with this manuscript and the url link to figshare is dead. This makes it quite hard for me to understand the complexity of your results.

 

Lines 273-293: These results strictly relate to descriptive statistics for the predictors and response, right? You seem to then do a bunch of comparisons of these predictors to one another conditional on other predictors. Are these formal tests (like t-tests?) or are you just visually comparing them (like 285 for example, what is the +/- part you are reporting here). Same for 286-293 with the comparisons across streams. I would separate the circuitscape model from this descriptive statistic/exploratory section and describe the results more in full.

 

In Table 1, stream is a factor not each stream individually. This factor has 7 levels resulting in 7 predictors, one for each stream. You don't need to report these p-values here for the differences between all other streams to Camp (the intercept and level 1 assuming you are using the constraint where level 1 for stream is set to zero), as this seems an arbitrary baseline (first alphabetically). The pairwise comparisons that are in the Supplement are more interesting and relevant (309-314) so you could move those here. Alternatively, switch to the sum to zero side constraint (contr.sum) and then the stream parameters would reflect differences between each and the grand mean, except for the last one, or just report the F-test.

I would discuss the results for the 3 suites of models first in the results before moving to the NDMS results. This seems like it is a better grouping given the way you have structured your hypotheses.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This study analyzed the community structure of pests and pathogens using willows (Salix sitchensis) as their host plant species. As far as I understand, willows represent the primary successional tree species establishing after a major catastrophic event: the eruption of Mount St. Helens. The study showed that landscape connectivity structures community composition of natural enemies (insects and one fungi) while other studied factors such willow populations distance to stream, leaf chemistry and leaf area showed varied effects on natural enemies` communities. such as the invasive stem-boring poplar weevil (Cryptorhynchus lapathi). Their findings indicate that landscape configuration is a strong driver of herbivore richness in primary successional forests. The authors suggest that differences in landscape connectivity among willow populations may help to promote ecosystem function by mitigating the impact of an invasive pest species (C. lapathi).The sampling methods and data are well designed and analyzed. Nevertheless, it was not clear from the Introduction that the study is focused on the natural enemies (pests and pathogens) attacking  S. sitchensis populations in different watersheds. I suggest changing the title to something like:  “Spatial habitat structure assembles herbivore communities in willow populations across the primary successional watersheds of Mount St. Helens“.

 

In my opinion, the manuscript should be more focused on the impact and importance of natural enemies in forest succession at different spatial scales (e.g., local, landscape, watershed) considering the low mobility of the willow`s natural enemies. Please, check this reference regarding the “scale of effect”.

 

Moraga AD, Martin AE, Fahrig L. The scale of effect of landscape context varies with the species’ response variable measured. Landsc Ecol. 2019; 34: 703–715. https://doi.org/10.1007/s10980-019-00808-9

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This is indeed an interesting article, with useful insights. My methods/results/discussion/conclusion is appropriate and acceptable. I have a major concern about the introduction, as it currently lacks enough context/background information, and reads more like a "Short communication" than a "Full-length research article".  Please revise accordingly, also the last paragraph needs more refinement. 

Below are some specific comments/suggestions for your consideration:

In the title, please add country/region as well (USA/Washington);

Page 1, abstract, please provide a few lines about the context/importance of this study and major quantitative findings;

Page 1, key-words: please revise, consider the search engine optimization (common keywords used by people to locate relevant research);

Page 3, lines 119-135: here it tells mostly about hypothesis and assumptions, ideally, it should also discuss what the author(s) have done to prove/justify their hypothesis/assumptions, and how this will contribute to science and/or forest management across scale (USA and beyond);

Page 10, Figure 3: please revise the size of the font'/legends (too large now!);

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

I have had the opportunity to review an earlier version of this manuscript, and in the revised version authors have adequately addressed most of the comments and/or suggestions.