Urbanization and agricultural intensification destabilize animal communities differently than diversity loss

Despite growing concern over consequences of global changes, we still know little about potential interactive effects of anthropogenic perturbations and diversity loss on the stability of local communities, especially for taxa other than plants. Here we analyse the relationships among landscape composition, biodiversity and community stability looking at time series of three types of communities, i.e., bats, birds and butterflies, monitored over the years by citizen science programs in France. We show that urban and intensive agricultural landscapes as well as diversity loss destabilize these communities but in different ways: while diversity loss translates into greater population synchrony, urban and intensive agricultural landscapes mainly decrease mean population stability. In addition to highlight the stabilizing effects of diversity on ecologically important but overlooked taxa, our results further reveal new pathways linking anthropogenic activities to diversity and stability.

Overall the findings are interesting and an important contribution to understanding large-scale patterns in community dynamics. I encourage the authors to provide more clarity in terms in order to speak to a broader readership. A table of definitions (ecological and statistical) would be helpful.
The paper could be improved by drawing more on natural histories of the species studied to inform analyses.
I appreciate that the authors were thorough in analyzing data with different landscape buffer sizes, but the landcover data was from 2006 and the agricultural intensity data from 2006-2013, and I wonder how failing to meet assumptions of static landscapes affects the study. The authors could also make it more clear the composition of the focal habitats for each species community.
The citizen science projects are exemplary programs that provide reliable data. I'm concerned about the use of BBS data for the purposes of the paper because the BBS protocol involves fixed dates over years and the data were largely influenced by fall and spring migrants which may have shifted the timing of their migration in response to climate change.
For a paper examining mechanisms of community dynamics, I find it a limitation that communities were examined within taxonomic silos. Species depend on species in other taxonomic groups, so why would we expect the diversity of one taxonomic group to be of high importance in affecting the stability of that same taxonomic group? Related, the authors are vague on the focal habitat type for the taxonomic groups studied ---do they all share the same habitat?
A confusing point for me relates to how (or when) the surrounding landscape was measured. The authors use landcover data from 2006. I've never been to France and so I don't know if the landscape is no longer being converted into new landuses, but a static landscape seems an unlikely assumption. Could the landscape really not have changed over the years of the study (6, 11, 17 years)? The authors also use a value related to the mean agricultural spending between 2006-2013 as index of agricultural intensification, though citizen science data on species abundance were collected through 2018. With the landscape data, the authors measure habitat conversion around focal sites and make implicit assumption about habitat conversion leading to habitat degradation. They should make those assumptions explicit, or confine inferences to the effects of the surrounding habitat conversion (which is related to habitat loss, connectivity, and other landscape features in addition to potential degradation of habitat that remains). Specifically, they assume urbanization is a detriment to semi-natural and agricultural areas while agricultural intensification is a detriment to woodland and semi-natural areas. Why wouldn't agricultural intensification, like urbanization, also be a detriment to agricultural areas? While at least some details are provided about the buffer habitats, I could not find details about the focal habitats. For example, for the bat samples, how many local pops were in each of the different land cover types? For bird samples, how many BBS plots were distributed in which habitat types? Ditto for butterfly garden counts. For a paper examining mechanistic aspects of community dynamics, the authors are vague on natural histories of the taxon studied, and those natural histories are necessary to make sense of their results. How would it affect the results if some butterfly species had metapopulation dynamics in which "stability" manifests as localized boom-bust dynamics followed by re-colonization events?
With regard to the citizen science projects, the three selected have standardized protocols and long-term data which are two features that favor data quality. As the authors point out, the bat survey is biased for roads with low traffic. I could not find information about the focal habitat for the 152 local bat communities studied. The project requires quite a bit of work from volunteers which leads me to strongly suspect the competent volunteers remain and provide reliable and trustworthy data.
The Birding Bird Survey is 2x2km2 plots randomly selected in which 10 survey points to cover range of habitat types within a given plot (but again, not clear details about the proportions of these habitats in the final sample). The BBS involves sampling twice per year on exactly the same dates per site per year, during fall and spring migration which means that these data do not represent the local community of birds, but migrants passing through. Again I suspect this intense project attracts highly competent birders capable of providing reliable and trustworthy data. However, the timing of avian migration peaks have been found to change in response to climate change in some areas, and with a large component of abundance being migrants, it raises question about whether trends in abundance reflect local changes or changes in timing of migration for these 269 bird communities -which they call "local bird communities" but not clear to me whether they excluded migrants.
The Butterfly data come from "gardens" which are residential habitat. The authors are appropriately conservative in restricting the butterfly/month species because identification by volunteers for some is difficult.
Reviewer #2 (Remarks to the Author): The major claims of this paper is that human land use and species diversity have similar effects on the destabilization of animal communities. The relationship between human impacts on diversity and community stability has been a topic of study for decades in ecology. As such, this paper is on an important topic. However, what the authors present is not groundbreaking. They do look at stability and diversity in a couple of ways, but there is no mechanism. It is a nice empirical study, but I do not know what more it provides than, for example, Blüthgen, Nico, et al. "Land use imperils plant and animal community stability through changes in asynchrony rather than diversity." Nature Communications 7 (2016): 10697. Comments 1. The authors performed a regression on of weighted MPD on SR. Why was a phylogenetic metric that includes species abundances chosen? What is the relationship between unweighted MPD and community stability? Including abundances into an explanatory variable introduces statistical independence biases (i.e., including an independent variable that derived from the dependent variable). The residuals (see figure S15) may certainly help reduce independence biases but then one must assume that the regressions that produced the MPD residuals well fit the data. If the authors want to use the weighted metric, then instead of richness they should use a diversity metric instead. Regardless of choice, the authors must include the unweighted MPD metric as this is more analogous to the SR metric and is the metric most often used. Note that the expectation of unweighted MPD is not dependent on SR (the actual values are but the expectation is not), thus it is less likely that the authors would need to The authors are using monitoring data on birds, bats and butterflies collected by citizen scientists to assess community stability over time. However, the time series for bats is quite short for this types of analyses. According to the authors (lines 290-291) the bat data consist of time series on only 4-6 years. This is very short in an ecological context, and is unlikely to include the full variability of bat communities. Hence, it is likely that they are under-estimating the community variability. The butterfly data does not consists of surveys of the full community. Instead, citizen scientists are recording species on a "closed list of 28 common species or species groups" (line 278), of which only 14 were used in the analyses. This is probably far from the full set of species encountered over a period of 7-11 years, even in urban environments. The authors state that "the temporal variability of the butterfly community restricted to these 14 species reflects that of the entire community" (lines 282-283). It is however unclear if they by "the entire community" here actually refer to the 28 common "species or species groups", or if they actually refer to the full community. Since they probably only have data on the 28 species or species groups I assume that this is what they refer to. If so, I find it misleading to say that "the temporal variability of the butterfly community restricted to these 14 species reflects that of the entire community". And more importantly, since they do only sample a subset consisting of common species, I find it likely that they are under-estimating the community variability. For these reasons, there is a considerable risk that the authors are under-estimating the variability of both bat and butterfly communities, with unknown consequences for the main conclusions.
Further comments on the data: -Butterflies were monitored in residential gardens. Yet most of the monitored species are unlikely to actually breed in many of these gardens, but rather use than for foraging only. So, if only a part of their habitat is monitored it is unclear what the inter-annual variability in community composition actually means.
-Also, the monitored species include two migratory species (Vanessa atalanta and V. cardui), of which at least V. cardui is unlikely to overwinter in France, and the population variability depends mainly on conditions in North Africa and the Mediterranean region.
Other major comments: In the abstract, the authors state that both population synchrony and population stability have effects on community stability, but through different drivers, i.e. are important, but that diversity loss is mainly affecting synchrony and habitat degradation mainly decreases population stability.
Yet, in the main text the authors mainly discuss the effects of habitat degradation on stability, and very little about diversity and synchrony. This inconsisteny needs to be resolved.
The different environmental drivers are estimated at different spatial scales. Landscape variables are assessed within radii of 250, 500 and 1000 m for bird and butterflies and in squares of 2-3 km for birds. In contrast, agricultural intensity variables are assessed with administrative regions (of unclear size, but probably much larger than the areas for which the landscape variables area assessed). This means that it is problematic to conclude that "while butterfly community stability is mainly impacted by urbanization, bird and bat communities are mostly destabilized by agricultural intensification…". In reality, it might instead be that butterfly communities are impacted by (any) drivers at smaller spatial scales while bird and bat communities are impacted by (any) drivers at larger spatial scales. Also, the radii of 250-1000 m seems quite small for butterflies an especially for bats, which are flying over rather large Minor comments: -I would like to see something about methods or the type of data used in the Abstract.
-Lines 285-286: "…gardens that had been monitored in July…" Does this meant that only data from July was included in the analyses? But several of the butterfly species are active mainly during other parts of the year?? -Line 377: Why did you use Corine Land cover data from 2006, and not more recent data (which I believe should be available).

2
This paper examines data from three citizen science projects (birds, bats, butterflies), to see 3 whether species diversity (within taxonomic groups) and habitat degradation (conversion of 4 surrounding habitat) affect community stability (within same taxonomic groups). Similar 5 research has been carried out with plant communities and the authors consider the lack of such 6 studies on animal communities to be an important research gap to fill. The author use structural 7 equation modeling to tease apart pathways affecting community dynamics. They report that 8 species diversity affects the synchrony of populations and habitat degradation (conversion of 9 surrounding landscapes to agriculture and urban areas) affects the stability of populations.

16
We added a supplementary table with ecological and statistical definitions in the manuscript 17 (supplementary Table 2):

19
Definition Equation Temporal stability of community total abundance Reflects the amplitude of inter-annual variations in the total abundance of the community in a site.

Weighted mean population stability
Mean amplitude of inter-annual variations of population abundances in a community, weighted by their relative abundance.

=
Population asynchrony Reflects the negative correlation degree among species temporal abundance fluctuations in a community.

= (∑ )
Shannon diversity Species diversity measure that takes species richness and evenness into account. In this study, we computed for each community the exponential of the Shannon index based on the total community species richness and evenness.

Species richness
Total number of species seen at least one year during the time series. Species richness is computed for each site and each dataset. Weighted or non-weighted Mean Pairwise Distance (MPD) Reflects the phylogenetic diversity inside a community. The standard MPD is computed as the mean pairwise phylogenetic distance between species in a community. The weighted MPD is computed the same way, but distances are weighted by the product of the abundances of the species that constitute the pair Footnote: With CV the coefficient of variation of community abundances, µ the community 20 mean abundance, σ the standard deviation of community abundance, the weighted mean 21 coefficient of variation of population abundance in a community, the mean abundance of 22 the population i in a community and its standard deviation, the synchrony between 23 population abundance fluctuation in a community, H' the Shannon index, n the species of a 24 community and the proportion of the abundance of the species i in its community.

30
We agree that the natural history of the species studied can inform our analysis and help to interpret 31 the results. However, we also think that the main strength of the paper is to show common patterns among taxa with very different natural history and that highlighting too much the peculiarities of each 33 taxa may blur the main message. To account for this point, we made the following changes in the main

51
The reviewer raises an important point. We agree that while the landscape context, i.e., the amount of 52 degraded land around the study sites, can affect the stability of animal communities, its variation 53 during the monitoring period could also have an effect.

55
To assess the extent to which landscape modification occurred in our monitored sites throughout the 56 study periods, we looked at changes in the land cover using the Corine Land Cover database that is

67
Our analysis indicates that more than 99%, 76% and 86% of the monitoring sites experience less than 68 10% change in the landscape surrounding them, for bats, birds and butterflies, respectively. Therefore, 69 for a large majority of our sites, the hypothesis of static landscape around our study site does not seem

83
We now present these new analyses in the main text and in the supplementary materials. Please note 84 that our results were mostly unchanged by these modifications of the landscape variables in our 85 analysis (see Table below

93
Except for the butterfly monitoring scheme, the monitoring programs do not target specific habitats.

94
Instead, observers were given random locations where to monitor bats or birds. The idea being that all

103
To clarify this point, we added the following sentences to the description of the protocols:

105
"These circuits were chosen to be close of the volunteer residency, with low-traffic roads for security 106 and representative of the different land-cover types in the area."

108
"Plots are squares of 2x2km 2 randomly selected by the national coordinator, within which the surveyor 109 places 10 points separated by at least 300m, in order to cover all the habitats present in the plot."

111
"Participants identify and count Lepidoptera in their own garden."

113
See also our response to comment # 8 of reviewer 1 where we clarified the way landscape context is

129
Previous studies found that bird migrants arrived at their breeding grounds on average 2 days per

144
We agree with the reviewer that analysing the three communities together would be excellent.

145
However, the data comes from three different citizen science programs; each one focalised on a given 146 taxonomic group. As a consequence each site provides data for only one taxonomic group, making it 147 impossible to analyse the effect of one taxonomic group on the others.

149
We modified the main text to make sure that readers understand that each site contains only one 150 taxonomic group counts, by adding the following sentence line 55-57:

151
"The three taxonomic groups were not monitored on the same sites as data come from three 152 independent citizen science programs."

154
Regarding the focal habitat, see our response to comment # 4 reviewer # 1 and response to comment #

165
As outlined in our response to comment #3 of reviewer 1, we now include additional land cover data

190
We believe that there is a misunderstanding here. We do not measure habitat conversion but landscape

210
Regarding the focal habitat, see our response to comment # 4 reviewer # 1 and response to comment #

220
See our response to comment # 2 reviewer # 1 regarding the natural history.

222
The use of nationwide dataset on natural community, necessarily implying correlative approaches,

223
prevents us to test for all possible mechanisms. Such dataset are necessarily noisy but still can reveal 224 general patterns that can be related to what is expected from theory or can generalized results found in 225 smaller but more controlled systems.

227
We agree that metapopulation dynamics might be one of the mechanisms underlying the dynamic of 228 some species included in the datasets. However, we do not think that this should question our results.

229
If the population of such species exhibit boom-bust dynamics at the monitored sites, they will increase

240
Regarding the focal habitat, see our response to comment # 4 reviewer # 1 and response to comment # 241 3 reviewer # 3

243
There is indeed some turnover of volunteers in these schemes, some of them participating for one or

271
The French BBS does not include a survey date during fall migration.

273
We re-run the analysis with the Dataset 1 after removing the transaharian migrants, with the Shannon

288
The Butterfly data come from "gardens" which are residential habitat. The authors are 289 appropriately conservative in restricting the butterfly/month species because identification by 290 volunteers for some is difficult.

292
We thank you for you positive comment.

296
The major claims of this paper is that human land use and species diversity have similar effects

302
The use of nationwide dataset on natural community, necessarily implying correlative approaches,

303
prevents us to test for all possible mechanisms. Such dataset are necessarily noisy but still can reveal general patterns that can be related to what is expected from theory or can generalized results found in 305 smaller but more controlled systems.

307
We disagree with the reviewer regarding the lack of mechanisms. As presented in our study, there are 308 two pathways by which community stability can be affected, either by a change in the stability of the

327
We thanks the reviewer for his or her positive comment. The main difference between our study and

360
Finally, unlike Blüthgen's study, our analysis also includes phylogenetic diversity as a possible 361 determinant of community stability, with result supporting the hypothesis that species response 362 diversity to environmental variations is a key mechanism for population asynchrony in communities.

382
We agree with the reviewer that using analogous diversity metrics that either both include species 383 abundances or not is preferable. We thus rerun our analysis the two ways, either with the Shannon

386
We think that accounting for species abundance seems to make more sense. Indeed phylogenetic 387 diversity is expected to foster community stability through increased asynchrony because unrelated

398
Are the more diverse communities more likely to have rare species that were removed from the 399 analyses and does including these rare species influence the relationships?

401
Regarding the bat dataset, we did not use cut-off and all contacted species were included.

403
Regarding the butterfly dataset, we did not use cut-off neither.

418
The reviewer is right about the fact that more rare species were found in the communities with the 419 more species in the dataset with 75 species (see below).

425 426
To assess the impact of excluding rare species, we re-run the analysis with the full set of species,

427
including 234 bird species. The results were qualitatively similar to the ones obtained with the 75 most 428 common species (see comparison in the two tables below).

455
To check for the potential multi-collinearity in our models, we performed variance inflation factor 456 analysis for the models presented in the main text, i.e. the Dataset 1, the spatial scale with the best AIC 457 and for both Shannon index / weighted MPD and species richness / non-weighted MPD. All our 458 variance inflation factor are less than 1.5, suggesting the absence of multi-collinearity in our models.

475
Although we agree that a table might not be the most efficient visually, we think that presenting those 476 numbers with a graph might not be the most appropriate either. Table 1 summarizes at the path level 477 the effects presented graphically in Figure 4. It allows comparing the various paths for the three 478 studied taxa in, we think, a very efficient and concise way. Consequently, we prefer to stick to a table.

479
However, we took this reviewer idea to present the results of the robustness analysis in the

482
For Figure 1, we log transformed y axes.

484
For Figure 2, which is now Figure 3 in the main text, we plotted all relationships and we plotted z 485 scores of species diversity and phylogenetic diversity to improve readability.

505
Note that the objective of our study is to assess the impact of diversity loss and habitat degradation on 506 the community stability and not to quantify the full variability of the studied communities. Since the 507 length of the time series we used does not vary with the diversity of the studied communities nor with any of the two landscape gradients, there is no reason to believe that the length of the time series could 509 bias our results.

510
The reviewer does not either suggest that longer time series would affect differently richer 511 communities or communities surrounded by intensive agricultural land, urban areas or more natural 512 landscape, which would be a problem.

514
Finally, we do not agree that the length of our time series are too short to assess community stability 515 nor that they are particularly short regarding other studies on the same topic. The mean life expectancy 516 averaged across species included in our analysis is 2.9 years for bats (see Table 1 for species level 517 estimates and references), which is under our time series lengths. Hence, for each taxon, our time 518 series are sufficient to capture community stability over several generations. Also note that compared 519 to classical studies on diversity-stability relationship in plant communities, the ratio of the length of

559
We agree that the use of "entire community" was misleading and we replaced it by "the community 560 including the abundance of the species groups".

562
To investigate the potential consequences of not sampling the full community on the estimation of the 563 community stability we used some extra-data. In addition to the 28 species and species groups, the 564 volunteers can also use an additional list with three species groups and three species of Rhopalocera 565 and six Heterocera species. However, the use of this extra list is not mandatory and actually few 566 participant use it, although we don't know if it is because they don't use it or because they do not 567 observe the corresponding butterflies. We recalculated the community variability with the 28 species

580
This high correlation between the three subsets is not surprising as rare species contribute less to the 581 variability of the community as their abundance is low. Note that contrary to the expectation of the 582 reviewer, we tend to overestimate community variability by restricting the data to the 14 species. But 583 as explained in our response to comment # 1 of reviewer # 3, the objective of our study is not to 584 provide the true value of community variability but to assess how variability is affected by community 585 diversity and landscape quality.

587
Since (i) we need species level identification to perform our analysis, (ii) we cannot be sure that all 588 volunteers use the extra list and (iii) the community variability is highly correlated among the three 589 datasets, we are confident that using the dataset on the 14 species does not bias the result of our 590 analysis and insure that our dataset is homogeneous in term of sampling.

592
Regarding the impact of including rare species on our results, please see our response to comment # 3 593 of reviewer # 2.

595
#3. Butterflies were monitored in residential gardens. Yet most of the monitored species are 596 unlikely to actually breed in many of these gardens, but rather use than for foraging only. So, if 597 only a part of their habitat is monitored it is unclear what the inter-annual variability in 598 community composition actually means.

600
All the 14 butterfly species studied forage in gardens but we agree that we do not know if they actually

614
We agree that the population variability of migratory species depend on conditions that we do not 615 account for in our analysis. However, the local conditions we analysed might still play a role by 616 affecting the survival of individuals. Further, if migratory species, which population variability are 617 driven by conditions outside our study area, were fully driving the dynamic of the studied 618 communities, we would expect that to blur the signal and not to find any effect of the landscape 619 characteristics surrounding our study sites. So including these migratory species is conservative 620 regarding the result we find.

622
However, to test the effect of these migratory species on our results, we re-run our analyses removing 623 V. atalanta and V. cardui (see below). The results obtained are qualitatively similar, highlighting the 624 same relationships between habitat degradation, community diversity and community stability.

625
Exclusion of these two migratory species mainly lead to the detection of stronger negative effects of 626 habitat degradation on butterfly community stability, suggesting that observed effects of landscape 627 characteristics might indeed be partly attenuated by the presence of migratory species.

651
We thanks the reviewer for this remark. We expanded our discussion of the relationships linking In this revision, the authors addressed my comments and concerns. I appreciate this paper looking at the mechanisms by which landscape changes affect animal communities.
Reviewer #2 (Remarks to the Author): The authors have presented a well written and analyzed study on what determines community instability. The authors well addressed my previous comments.
Reviewer #3 (Remarks to the Author): The revision has significantly improved the quality of the manuscript, but in my view there are still some outstanding issues. I maintain that I think that the time series is short for analysing community stability over time.
Invertebrates such as butterflies are known to fluctuate at least an order of magnitude in population size between years, and this also affects community dynamics. The argument that other studies also have used short time series does not solve this issue.
My previous comment that only a part of the habitat was sampled (for butterflies) is also not fully resolved. The authors argue that population declines should be reflected in gardens even if butterflies only forage (and not reproduce) there. But the point is that the extent to which butterflies use gardens for foraging might differ between years, depending on e.g. weather. In a dry year, an irrigated garden might be used to a much higher extent than natural habitats, but this might not be true in another year.

After careful consideration, my colleagues and I agreed that the two outstanding concerns raised by
Reviewer 3 on the study duration and the representativeness of the habitat sampling for butterflies have been partly addressed already. In particular, we are inclined to agree that the effects of "boombust" dynamics of single species should be dampened in your community-level analysis, as you argue in the response letter. However, it is important that this point is more clearly expressed in the main text, along with other caveats on potential issues regarding 1) the duration (for all taxa), and 2) potential biases of the sampling scheme (especially for the butterflies).
To clarify the limitations of our study, we added a paragraph lines 174 to 190 in the "limit and conclusion" section part, discussing the impact of the time series duration, in particular in relation to the estimation of population and community variability, and the potential biases related to the butterfly sampling scheme:

"Limits and conclusion
Here we measured community stability at a relatively short-time scale (up to 6, 17 and 11 years for respectively bat, bird and butterfly communities), reflecting the time scale used in most studies on the relationship between diversity and community stability 6 . However, population and community temporal variability are known to increase with the considered time scale 40-42 and as such, our estimates of temporal variability might underestimate the full variability of the studied communities. While this should not affect the effects of landscape composition we found, and indeed our results are robust when compared with analyses on two subsets of our datasets with different time series durations (see Methods and Supplementary Fig. 7-9), longer time series would be required to estimate the full variability of the studied communities. Another limitation of our study is that we assessed habitat degradation at the landscape scale and did not account for local conditions, such as management practices, that could also affect community variability. For example, butterfly data were collected in private gardens with different management strategies that are known to affect the attractiveness for butterflies 43 . Accounting for such management practices as well as other local scale characteristics such as habitat heterogeneity that is also known to affect population stability 16 would improve our understanding of the determinant of community stability." We also ask that you take special care to ensure that statements throughout the manuscript do not inflate the strength of the results, and that the terminology with respect to population and community stability is not misleading, e.g. the statement in L157-159 of the manuscript with tracked changes suggests that population stability of single species was measured separately from the community data rather than partitioned from the latter using the approach from reference 22. (Please note that the Methods section does not contribute to the word limit, and therefore could and should be expanded as needed to ensure that readers can follow it.).
We thank you for raising this issue. We carefully checked the terminology with respect to population and community stability, adding "weighted mean" to "population stability" to make clear that we talk about the component of the community stability used in the statistical analyses. This change was performed lines 12, 78, 114, 123, 125. As explained above, we added a full paragraph about the limitation of our study and further made some change to not overstate our results, for example lines 100 and 145-146.
Finally, we ask that you clarify why the data and codes are only available upon request and were not shared for the review process. At a minimum, we ask that you provide the raw data underlying the key figures, please see the "Data Deposition" and "Source Data" paragraphs below.
We modified the data and codes availability in the manuscript, and we uploaded our data and codes in a Zenodo repository [https://doi.org/10.5281/zenodo.3678366].
In addition to these changes, during the revision process, we noticed several typos that we corrected as well as some unclear parts that we tried to clarify. This includes typos in the values presented in fig.3; some clarifications of the axis labels in the supplementary figures and in the legends of the figures; a clearer explanation of the effects of landscape composition on population stability.

Reviewer #3
I maintain that I think that the time series is short for analysing community stability over time. Invertebrates such as butterflies are known to fluctuate at least an order of magnitude in population size between years, and this also affects community dynamics. The argument that other studies also have used short time series does not solve this issue.
See our answer to editor's comment above and the insertion made in the limit section of the manuscript, line 175-184: "Here we measured community stability at a relatively short-time scale (up to 6, 17 and 11 years for respectively bat, bird and butterfly communities), reflecting the time scale used in most studies on the relationship between diversity and community stability 6 . However, population and community temporal variability are known to increase with the considered time scale 40-42 and as such, our estimates of temporal variability might underestimate the full variability of the studied communities. While this should not affect the effects of landscape composition we found, and indeed our results are robust when compared with analyses on two subsets of our datasets with different time series durations (see Methods and Supplementary Fig. 7-9), longer time series would be required to estimate the full variability of the studied communities." My previous comment that only a part of the habitat was sampled (for butterflies) is also not fully resolved. The authors argue that population declines should be reflected in gardens even if butterflies only forage (and not reproduce) there. But the point is that the extent to which butterflies use gardens for foraging might differ between years, depending on e.g. weather. In a dry year, an irrigated garden might be used to a much higher extent than natural habitats, but this might not be true in another year.
See our answer to editor's comment above and the insertion made in the limit section of the manuscript, line 184-190: Another limitation of our study is that we assessed habitat degradation at the landscape scale and did not account for local conditions, such as management practices, that could also affect community variability. For example, butterfly data were collected in private gardens with different management strategies that are known to affect the attractiveness for butterflies 43 . Accounting for such management practices as well as other local scale characteristics such as habitat heterogeneity that is also known to affect population stability 16 would improve our understanding of the determinant of community stability.