Tree loss impacts on ecological connectivity: developing models for assessment.

Trees along linear features are important landscape features, and their loss threatens ecological connectivity. Until recently, trees outside of woodlands (TOWs) were largely unmapped however; the development of innovation mapping techniques provides opportunities to understand the distribution of such trees and to apply spatially explicit models to address the importance of trees for connectivity. In this study, we demonstrate the utility of models when investigating tree loss and impacts on connectivity. Specifically, we investigated the consequences of tree loss due to the removal of roadside trees, a common management response for diseased or damaged trees, on wider landscape functional connectivity. We simulated the loss of roadside trees within six focal areas of the south east of the UK. We used a spatially explicit individual-based modelling platform, RangeShifter, to model the movement of 81 hypothetical actively dispersing woodland breeding species across these agriculturally fragmented landscapes. We investigated the extent to which removal of trees, from roadsides within the wider landscape, affected the total number of successful dispersers in any given year and the number of breeding woodlands that became isolated through time. On average roadside trees accounted for less than 2% of land cover, but removing 60% of them (~1.2% of land cover) nevertheless decreased the number of successful dispersers by up to 17%. The impact was greatest when roadside trees represented a greater proportion of canopy cover. The study therefore demonstrates that models such as RangeShifter can provide valuable tools for assessing the consequences of losing trees outside of woodlands. impacts on ecological developing See general comments. The paper does not explicitly study the impacts of tree disease, but rather the impact of tree loss. Similarly, the paper does not talk or develop scenarios about mitigation strategies, it’s just a possible output (word) that appears in the title, abstract and conclusion. I would suggest to find a title oriented toward the We have amended the title to ‘Tree loss impacts on ecological connectivity: developing models for assessment’. Abstract See my general comments for the direction given to presentation of the study. Context: it is unclear from this section that the focus is on animal connectivity favored by trees. I’m not sure of the interest of the “tree disease” focus; it could more simply be just tree loss (by disease or management choice). [AM1] Objective: There is not true “response to ash dieback” (L19) taken into account in the study. Information about under study missing: plants, which Methods: The reference to Arcgis is not needed. The random selection for tree removal should be mentioned. I like an idea the of the 6 areas. Conclusions: The reference to “ mitigation strategies” is not supported by the study. We have altered the abstract to reflect more accurately the direction of the study and the focus on tree loss and connectivity rather than tree disease. Abstract Trees along linear features are important landscape features, and their loss threatens ecological connectivity. Until recently, trees outside of woodlands (TOWs) were largely unmapped however; the development of innovation mapping techniques provides opportunities to understand the distribution of such trees and to apply spatially explicit models to address the 14 importance of trees for connectivity. In this study, we demonstrate the utility of models when 15 investigating tree loss and impacts on connectivity. Specifically, we investigated the consequences of tree loss due to the removal of roadside trees, a common management 17 response for diseased or damaged trees, on wider landscape functional connectivity. We 18 simulated the loss of roadside trees within six focal areas of the south east of the UK. We used a 19 spatially explicit individual-based modelling platform, RangeShifter, to model the movement of 20 81 hypothetical actively dispersing woodland breeding species across these agriculturally 21 fragmented landscapes. We investigated the extent to which removal of trees, from roadsides 22 within the wider landscape, affected the total number of successful dispersers in any given year 23 and the number of breeding woodlands that became isolated through time. On average roadside 24 trees accounted for less than 2% of land cover, but removing 60% of them (~1.2% of land 25 cover) nevertheless decreased the number of successful dispersers by up to 17%. The impact 26 was greatest when roadside trees represented a greater proportion of canopy cover. The study therefore demonstrates that models such as RangeShifter can provide valuable tools for assessing the consequences of losing trees outside of woodlands.


1.
The manuscript in its current form is quite difficult to read. There seems to be some information missing, and then some information that feels superfluous. I have made specific recommendations below.
We have addressed the recommendations of all three reviewers and hope that along with the reframing of the manuscript it is easier to follow.
2. I am unsure of the novelty of the application. This seems more of a reframing of the use of RangeShifter in landscapes of differing compositions, but with roadside trees randomly removed to change the focus to tree disease. I feel the paper would be more effective if it were coupled with models of disease spread to create realistic patterns for analysis. This is suggested as a future avenue of research in the discussion, but I do think this is where the novelty would be. The authors find that differences between replicates account for up to 30% of the variationsuggesting that spatial structure is important for the conclusions. Therefore I would suggest that analysing realistic patterns of tree loss (as opposed to random) would make more sense.
We have followed reviewer 3's recommendation and changed the manuscript to focus on tree loss and connectivity rather than ash dieback per se. Thus we have reframed the paper as assessing the importance of trees as elements for connectivity, under different threats and the use of Rangeshifter as a tool to do so. We do not combine another model to locate sick trees and ash dieback spread to create patterns of tree loss as this would now be beyond the context of the revised manuscript. Furthermore given the multiple threats to trees, disease, climate and management, it would be difficult to determine how 'realistic' patterns will look, and we therefore keep the random approach for this initial study of tree loss and connectivity. We do however continue to highlight in the discussion the potential coupling of models of disease spread to identify patterns of tree loss to models analysing connectivity as an avenue for future work. Furthermore, trees along infrastructure features such as roads, railways and watercourses occupy an increasing proportion of all trees outside of woodlands, but the impact of tree loss on wider landscape connectivity, due to felling in response to tree disease or climate induced dieback, remains unexplored. Until recently, trees outside of woodlands were largely unmapped however, with the development of innovation mapping techniques there are now opportunities to explore the importance of such trees. Thus the novelty of our study is that we use a spatially explicit individual-based model (typically connectivity studies hitherto use approaches based on graph theory) which utilises innovative high resolution mapping data to consider the impact of the loss of these trees on wider ecological landscape connectivity, as a first step towards understanding the most appropriate management and recovery response.

2.
It would be helpful to explicitly state the question(s) you are answering and your predictions/hypotheses in the last paragraph of the discussion.
We have altered the final paragraph of the introduction to outline our research question more specifically.

Are 10 replicates enough for the landscape and demographic replicates? It would be useful to get an idea of the distribution of values (for successful dispersers and patch isolation).
For all tree removal scenarios (20%, 40%, 60%) on all squares, demographic replicate, together with its interactions with the four varied factors, accounted for < 0.01% of the variance in the number of successful dispersers and isolated patches (Appendix A, Tables A3,A4,A5). Therefore, we believe that 10 demographic replicates are in fact more than sufficient. Landscape replicates indeed accounted for up to 30% of the variation, indicating that the spatial pattern of tree removal is important for connectivity, and we believe this an interesting point in itself. However, increasing the number of landscape replicates would be unlikely to alter our main results; indeed testing a great number of replicates in the initial stages of the study did not greatly affect the results.
Running an increasing number of replicates always comes at the expense of computing time (running the 81 species and all the replicates over 6 different landscapes already resulted in over 150000 simulations). We feel that our choice of 10 demographic replicates and 10 landscape replicates is sufficient to generate robust results while maintaining tractable computing timescales. Furthermore with 81 species, 6 tiles, baseline scenarios and three removal scenarios the distributions of replicates could be shown for a possible 1458 different combinations. Therefore rather than present the distribution of values for replicates, for ease of reading, we chose to present the mean, min and max proportion of successful dispersers/isolated patches relative to the baseline landscape for each tree removal scenario on each square.

4.
General comment on the discussion: the results are represented and interpreted, but with almost no connection to related literature. I suggest interpreting the results in the context of other studies.
This is a good point and we have now improved the first three paragraphs of the discussion to incorporate existing related literature when we interpret our results.

L14-15: What questions do you plan to address?
L24-25: What species groups are the theoretical species meant to represent?
We have altered the abstract to address the above two points.

L33: Be more specific about the type of model RangeShifter is.
We now write '…spatially explicit individual-based modelling platform, RangeShifter' in the abstract line 18.

Done.
L83: Add to the end of the SMS sentence "which is embedded in RangeShifter" -it makes the introduction match up more clearly with the abstract and methods.

Done.
L78-98: It makes more sense to have this paragraph after the paragraph L99-120.
We have restructured the introduction following comments from all reviewers.

L122-126: This paragraph is out of place, and repeated later.
This paragraph has been removed.
L175: Change "those that fell out with" to "those that did not fall within the boundaries of" Done.
L179: What are the associated costs? Perhaps include these in the text and/or Figure 2 legend.
The costs can be found in Table 2, we have adjusted the text to direct the reader to table 2. L228-231: The bit in the brackets confused matters with the reading of the methods. I suggest moving to after "30 removal scenario landscapes" and changing to (10 replicate landscapes for each of the 20%, 40% and 60% roadside tree removal scenarios).
We have changed this.

L233: Why the 10 year burn-in? If standard provide a reference, if not provide justification.
The burn-in period is to allow the population dynamics within the model to stabilise. Burn-in periods vary depending on the model and simulations, and initial testing indicated that 10 years for sufficient for the simulation runs for the study. We have added a sentence to the methods to justify this.
L244: Which function/package did you use? Also, make sure to cite R and any packages used -it helps with reproducibility and also provides credit to package developers.
The package information and citations have been added.

L281: I make this 5% and not 9%
Thank you, this has been changed.
L283: I make this 0.7% instead of 3%. Also where are the results for HM and PR?
Thank you, this has been changed. For ease of reading we have included the main effects columns only. Interactions between factors were (with the exception of carrying capacity and per-step mortality risk for successful dispersers noted in the text) relatively unimportant and thus we chose not to present them.
L288: Given for each scenario and square the minimum change in number of isolated patches is negative and the maximum is positive, the mean is not really meaningful -yes the mean change is limited, but that's because some spatial configurations allow for a more substantial decrease and some an increase. Perhaps it's better to discuss the min/max in the results and not present the mean. This again provides an argument for showing the distribution of the values obtained for the replicates.
As highlighted above with 81 species, 6 tiles, baseline scenarios and three removal scenarios the distributions of replicates could be shown for a possible 1458 different combinations and thus for ease of reading we chose to present the mean, min and max. We have however changed the text to discuss the min/max now rather than the mean and highlight the reviewers point in L315. We also now discuss the positive and negative results in the context of other studies in the discussion L380.
L295: add "compared to the baseline" after "isolated patches" Done.

L296: Interactions are not shown in the table.
Similar to the table for successful dispersers; for ease of reading we have included the main effects columns only. Interactions between factors were relatively unimportant and thus we chose not to present them.
L297: Keep consistent with the rest of manuscript and change "DP" to "directional persistence". Done.

L361-378:
Where 'models' are mentioned to investigate the impact of tree loss on foraging habitat and shelter does this mean the same modelling approach? If so please explain how it could be applied. My understanding is that RangeShifter models dispersal in terms of emigration, transfer and settlement and I'm not sure how forage/shelter fits in to the modelling platform. If not, this paragraph should be removed or adapted.
We have removed this text as it was indeed confusing. Values >0.2 are highlighted in bold. We have added this to the figure legends.

Results for PR are missing from the table
We have added the PR column. Done.

What does the residuals column represent?
We have removed the residual column as it was unnecessary.

Where is the interaction column?
For ease of reading we have included the main effects columns only. Interactions between factors were relatively unimportant and thus we chose not to present them. Done.

Figure 5 and 6 -perhaps include the standard errors as error bars.
The standard error bars are too small to be seen on the graph, the standard errors are however presented in the appendix tables.
The authors create 81 virtual species by considering all possible combinations of four factors (carrying capacity, perceptual range, directional persistence, unsuitable habitat mortality) at three levels. Their conclusions are based on the mean proportion of these 81 virtual species that successfully disperse. I am concerned that we have no information on the proportion of 955 species that use ash trees or the 45 species are that are assumed obligate on ash that have each of these 81 different factor combinations. It is possible that a large proportion of species have very similar combinations of these four factors -in particular the 45 that are ash-obligates. Similarly there may be factor combinations that occur extremely rarely in the real world. I think that it is therefore misleading to conclude that removing 60% of roadside trees could decrease the number of successful dispersers by up to 17% (Line 306). I think that this study shows instead that 17% of 81 possible combinations of four factors relevant to dispersal would decline. If none of the 955 species that use ash possess any of these factor combinations then it is possible that there would be no decline of species at all. Alternatively, if these combinations of factors are common in ashusing species the decline may be much more severe than predicted. We know how many of the species that use ash are birds, vascular plants, lichens etc. It ought therefore to be possible to include a rough idea of how the virtual species types created by the authors map onto the characteristics of real species.
Following reviewer suggestions, we have reframed the manuscript and thus removed the focus on ash dieback. We therefore no longer believe it necessary to map the virtual species on to species using ash trees. We agree with reviewer 2 that the results will be relevant to only certain species however, and we acknowledge this in the last paragraph of the discussion. However, we also highlight that there are insufficient data on the dispersive characteristics of woodland species (ash dependent or otherwise) and thus until such data become available it would be difficult to do so. Rather than include only some 'realistic' assumptions mixed with theoretical assumptions, for parsimony, we create entirely virtual species. This also has the advantage of allowing investigation of parameter space. If, indeed, future empirical work on quantifying dispersive traits in woodland species discovers that such species do possess traits (that are highlighted in this study) that may make them vulnerable to tree loss then this could provide the basis for management and act as an early indicator of risk.
I was also concerned that they only attempted to model "active" species, and many of the most at-risk ash associated species are poor dispersers (eg lichens) -specifically because they tend to get stuck in little habitat pockets and end up very range-restricted.
In this paper we have deliberately focused on investigating the potential impact that the loss of roadside trees might have on the connectivity of species for which trees forms a positive component of the matrix. Our focus is thus on species that have the capacity to, at least occasionally, disperse successfully between the patches of woodland that, for these species, we consider breeding habitat to be. The reviewer is correct that there are many species that have very poor dispersal ability. For these species, the trees outside of the woodlands are likely to provide key patches of habitat that can form stepping stones via which the species can maintain connectivity between woodlandsthough this is a different type of connectivity as it occurs over multiple generations.
There are quite a few unreferenced assumptions in the model that could potentially have quite a big effect on the results: eg. Why did the authors assume that species would only reproduce in "breeding patches"? There are, I am sure, a number of species that breed in roadside trees. I would like to see more justification for this and other assumptions, or at least to see them varied to see how strong an effect they have.
This relates to the point above. We have decided to restrict ourselves here to species that need a woodland patch for reproduction. For these species, the trees outside of woodland improve the permeability of the matrix. We have not focused on species for which single trees outside of woodland provide suitable breeding habitat. However, we recognise that for a set of species, individual trees will provide important habitat. In future work, we will extend our modelling to investigate this. It requires first some technical developments as this will substantially increase both the number of suitable patches of habitat on the landscape and the total population sizes, requiring greater computing power. We have edited the manuscript such that our current focus is clearer L160.
I feel that it is important that the authors include a second analysis where trees in the "breeding patches" are also reduced. They mention this as future work, but I think it would be interesting for this paper because of the potential for a strong interaction between a decline in the numbers of individuals and reduced landscape connectivity.
This is again a good point and represents an interesting topic that we want to address in the future.
Our justification for not doing this in the current work is that we are focussing on the targeted removal of diseased or damaged trees close to infrastructure in the event of a disease epidemic or climate induced dieback. Thus the loss of a percentage of trees from woodland is not relevant for this question. Again, we have added some text (L117, L396) to the manuscript to provide clarity on our choices in this current exercise, which really is focused on the impact of tree loss near infrastructure on wider landscape connectivity for actively dispersing woodland species.
Line 99-100: I disagree with the statement "trees outside woods (TOWs) seldom if ever are selfseeded." In my opinion the dramatic decline in hedgerow management since 1945 has resulted in a huge increase in self-seeded ash, sycamore and hawthorn in hedges growing into adult trees. There are very large increases in TOWs over this period which cannot be attributed to planting.
With the restructuring of the introduction the above text has been removed and thus the above point addressed.
Lines 100-102: "Instead, they exist because they have been deliberately placed or at least allowed to persist."…..in other words, planted or natural regeneration. I'm not sure if there are any other options, so this seems like a truism.
With the restructuring of the introduction the above text has been removed and thus the above point addressed.
"Unlike natural regeneration in woodlands, without human intervention the loss of TOWs marks a permanent decline in canopy." I think that the authors are trying to say that in woodlands, canopy gaps are often rapidly filled. Trees lost from linear landscape features are much less likely to be replaced quickly.
With the restructuring of the introduction the above text has been removed and thus the above point addressed.
Lines 177-179: "Woodland patches were defined as the breeding habitat for the study species and other habitat types (roadside trees, matrix trees, matrix habitat) formed the inter-patch matrix each with a habitat-dependent movement cost associated." I think, in this context, that "study species" is not F. excelsior but the 81 virtual species mentioned for the first time later on in this section. Please reword to avoid this confusion.
We have reworded this.
Line 343 they say that the numbers of roadside and lineside ash trees will run into billions -I think this is extremely unlikely, but in any case the authors quote the Tree Council figure of 27-60 million ash trees outside of woodlands in total, so they can't then have billions of roadside ash trees.
We have removed this sentence.
This paper is a useful demonstration of a modelling method but until more work has been done on the functional profiling of ash-using species it tells us little about the real-world impacts on ecological connectivity and appropriate mitigation strategies.

Tree disease impacts on ecological connectivity: developing models for assessment and mitigation strategies
Comments to authors

General comments
This paper aims to show the relevance of using a modeling approach to assess the impact of tree disease on functional connectivity. The idea of using connectivity modeling approaches to measure the importance of small landscape elements is interesting and fits in the scopes of the journal. My main concerns relate to the orientation of the paper toward tree disease, while for what I understand of the methods, the trees could fall because of disease, management or storm without changing the results. I understand that the question of the impacts of tree diseases in a hot topic, but it seems to me that the tree removal scenarios are not specific to the disease. By oriented the paper toward the "disease" aspect, I would have expected spatial pattern of tree removal related to the disease (for instance linked to the spreading of species or link to management choices in response to the disease). Here the removal of trees appears to be random.

Such a random pattern may be relevant to represent the spatial spread of the disease, but in that case the point has to be done and justify in the text. Personally (and I have no idea of the epidemiology of the ash dieback, and such information should be given in the introduction/methods) I would have guess that the disease is spreading from a host to neighbors, creating clusters of falling trees (see discussion L345-350). The emphasis on the "disease" aspect is thus confusing, as the reader try to see how it's taken into account in the analyses, but can't find it. Generally, the logic
To summarize, it seems to me that the methods are adapted to answer the question of the relevance of modeling approach to assess the impact of tree removal, but not to answer the question of tree disease impact on connectivity, which would have imply combining another model to locate sick trees and ash dieback spread. I may be wrong, and some justification of the method underlying the scenarios may allow overcoming this problem. If I'm right, then I suggest rewriting the introduction (and related sections in the methods) to focus on the importance of trees as elements for connectivity, under different threats, diseases being one of them. Such change would enlarge the scope of the paper to "importance of trees for connectivity", which is relevant and interesting. It would also better fit with the points developed in the discussion.

Overall, he paper is mostly well written and pleasant to read. Early information about the taxa under study (insects) is missing for a good understanding.
Thank you. Following reviewer 3's suggestions we have reframed the manuscript to focus on the loss of roadside trees in response to more general threats than just ash dieback. Trees along infrastructure features such as roads, railways and watercourses occupy an increasing proportion of all trees outside of woodlands, but the impact of tree loss on wider landscape connectivity, due to felling in response to tree disease or climate induced dieback, remains unexplored. Thus the novelty of our study remains unchanged in that use a spatially explicit individual-based model to consider the impact of the loss of these trees on wider ecological landscape connectivity, as a first step towards understanding the most appropriate management and recovery response.

Specific comments I have very few specific comments. The text is clear and well structured.
Title See general comments. The paper does not explicitly study the impacts of tree disease, but rather the impact of tree loss. Similarly, the paper does not talk or develop scenarios about mitigation strategies, it's just a possible output (word) that appears in the title, abstract and conclusion. I would suggest to find a title oriented toward the We have amended the title to 'Tree loss impacts on ecological connectivity: developing models for assessment'.

Introduction
See my general comments. If the authors decide to keep the "disease" focus, then I find the introduction too general and not focused enough on the ash dieback. Some references to other diseases could be removed, while most of the information included in the "Study system" section should be given here. Information about the epidemiology of the disease is missing.

L122-126: That should appear before (maybe after L109?)
Similar to the abstract we have altered the introduction to reflect more accurately the direction of the study and the focus on tree loss and connectivity rather than tree disease.

Study system
L129-141: Some of this should be included in the introduction (if the "disease focus" remains), to depict a better picture of the extent of the disease and give an idea of its potential impacts.

I miss here a description of (i) the natural mortality linked induced by ash dieback (and how long does it take for the tree to die) and (ii) information about how the disease spread. This information is important to understand the potential impacts and the management choices along
roads. If the probability that a sick tree will contaminate neighbors is high, then fell large portions is necessary to stop the disease. If sick trees die within a year or two, then keeping them alive a little bit longer by deciding not to fell them will just preserve connectivity for these 2 years, but will have contributed to the spread of the disease. I think that this part needs to be partly rewritten to make the rational clearer, i.e. the justification of why sick trees shouldn't be cut.

L152-158: For me that belongs to the introduction
We have followed the advice of Reviewer 3 and reframed the manuscript to focus on tree loss and connectivity rather than ash dieback, this we have removed the study system section from the manuscript.

L244-245: What varied factors?
We have added the names of the factors (perceptual range, directional persistence, carrying capacity, matrix per step mortality risk).

Appendices
Please do not use acronyms in the tables, there is room here to facilitate the reading.
The acronyms have been replaced in appendix A, there is no sufficient space in the tables of appendix B to replace the acronyms. [AM1]

Highlights
We model the removal of non-woodland roadside trees and the effects on wider landscape connectivity

125
Study landscapes 126 Our study landscapes consisted of six 10km x 10km squares in the south east of the UK (Table 1, 127 Figure 1). This region is a good example of an area with trees under threat; ash dieback is 128 prevalent within the region and is expected to cause the catastrophic loss of ash trees that  inter-patch matrix (Figure 2a).The percent of the tree cover for each square and the 164 composition of the tree cover (matrix, road side, or woodland trees) is given in Table 1.     Table A1) and in the number of isolated patches (Appendix A, Table A2). Therefore, counts of

259
Similarly, the number of isolated patches was averaged across all demographic replicates and 260 years, and the effect of tree removal was represented by the increase in the mean number of 261 isolated patches relative to the corresponding baseline simulation.

262
To account for all species simulations being run on the same 10 landscapes replicates (LR) for a 263 given removal scenario in a particular square, the data were fitted separately for each square to 264 linear mixed models in which landscape replicate was included as a random effect. The least 265 squared means for the four varied factors (K, HM, PR, DP) were extracted from these models 266 using R package lsmeans (Lenth 2016) to illustrate the main effects of each model parameter.

268
Successful dispersers 269 For each square, the mean proportion of successful dispersers declined as the percent of trees 270 removed increased (Table 3). In general, the reduction in successful dispersers due to tree 271 removal was less than 10%, but for some individual parameter and landscape replicate 272 combinations, the reduction in successful dispersers could be up to 17%. Removing roadside 273 trees also changed the dispersal trajectories of individuals and increased the frequency of 274 disperser visits to cells containing non-roadside matrix trees (Figure 4).     At only 20% roadside tree removal, the increase in patch isolation over baseline levels was very 313 limited, but larger increases in isolation were observed at higher levels of removal (Table 4, 314 Figure 6). Overall, the mean change was limited because some spatial configurations allow for a 315 more substantial decrease and some an increase in patch isolation. For example, in the worst-316 case scenario the maximum increase in the number of isolated patches was 3.9 above the 317 baseline (Table 4). However, in some cases tree removal also decreased the number of isolated 318 patches compared to the baseline, minimum values ranging between -1.2 and -2.1 (Table 4).

319
Increasing the per-step mortality risk led to larger increases in the number of isolated patches, 320 whereas increasing directional persistence resulted in smaller increases (Appendix B Table B2, 321 Figure 6). Main effects and their first-order interactions generally accounted for a small 322 proportion of the variance in the isolation metric, although the influence of mortality risk and 323 directional persistence increased considerably as the proportion of trees removed increased.