The Importance of Connected and Interspersed Urban Green and Blue Space for Biodiversity: A Case Study in Cork City, Ireland

Round 1

Reviewer 1 Report

General Comments

Overall I was very unclear what the goal of the research was or how it adds to our knowledge about urban bird ecology. As written the manuscript did not clearly convey whether it was about urban bird ecology, testing basic landscape ecology ideas, or simply an evaluation of where birds are located. The Introduction lacked a clear narrative about the field of study and what is known and where gaps in knowledge are. Moreover, there were no a priori hypotheses. Essentially the field of urban bird ecology was not discussed and the ideas that were are not really needed or novel ideas. In the case of Methods there are a number of shortcomings on experimental design, how bird data were collected, and inappropriate use of statistical methods. Ultimately, what is needed is a thorough read of the urban bird literature to identify gaps and laying out a priori hypotheses that the data you collected can address.

Specific Comments

L10. Here and throughout ms please separate these as urban green spaces and urban blue spaces to be consistent with literature. Also, please avoid use of acronyms and simply write out words.

L35. This is a bit confusing as we generally think of these types of spaces falling within urban ecosystems. While ecosystem is not scale explicit, I would argue for many of the green spaces that these are not in fact ecosystems. Also, this is a run-on sentence. I would thus suggest revising this section.

L37. Change environs to ecosystems. We consider these to be urban ecosystems.

L38-43. I would suggest deleting these sentences as your ms is focused on urban ecosystems.

L45. This is also important to city managers, conservation practitioners, etc. I would argue it is actually of more interest and importance to conservation and management and human wellbeing than it is to biogeography and landscape ecology.

L46. Can delete the definition of connectivity. Readers should understand this concept and if not, then maybe the article is better suited for a standard ecology journal.

L52-72. Again, I would suggest cutting most of these items. Don’t need to review what landscape metrics are and the issue of scale at which these are run has been well discussed in the landscape ecology and urban ecology literature. I think you can greatly simplify such points to be the type of metric used matters as does the scale of the analysis.

L73. I’m not clear the importance of this paragraph. If the issue is simply that we have better satellites today, that’s great, but to me this is mainly methodological. That said, you’ve cited papers that indicate we know well what can be included in green space assessments and we have good tools to find them, even small parks, etc.

L80. Actually this isn’t really the case. The paper you cite is 13 years old and the reviews you cited earlier in the ms are much more recent descriptions of what a green space is. For instance articles in the last five years are not including water in the term. I feel this paragraph is a bit of disservice to where the field is at currently. Ecologists very well know how to differentiate green and blue spaces. Notably, there may some misunderstandings here in that individual researchers define what they are evaluating in a given paper and certainly there are papers that lump terrestrial and aquatic together for certain types of assessments.

L90. What do you mean lack of consensus on what configuration matters? I don’t think there has really been much work on testing specific real world configurations against null or random configurations. Also, configuration would be species dependent, so I’m at a loss as to what you mean here.

L93-96. These questions really were unexpected. First, there is a rich literature on urban bird ecology that has been missed in the Introduction. Second, there is no clear establishment of what hypotheses need to be tested in urban bird ecology here. Third, the location of the study is not important as you should simply be using a location to test hypotheses. That is, you can move the location information to Methods.

L116. I’m unclear why you are building a land cover database for Cork rather than simply using an established land cover database? You have not set up any a priori hypotheses and predictions as to why a novel land cover database is needed. Also, your land cover information should include minimum mapping unit or pixel size, projection, and datum used. Finally, why use 2018 data when your field surveys were in 2019?

L144. For point counts are accounting for detectability? If not, then this needs to be done. Also, are you using a fixed radius point count and if so what is it. Lastly, how long were your point counts run for?

L146. What are your target group of birds? Migrants? Breeders? I have no clear idea where in the city you are surveying, what you are surveying for, and how this answers novel or important hypotheses in urban bird ecology.

L146. Did you only count a location once? Using only one observation in one season is not a valid approach to enumerate birds.

L148. How are you avoiding double counting?

L153. You need to describe an experimental design first and then indicate how you surveyed locations within the design.

L160. Why? You are working on birds and these don’t necessarily make sense as boundaries.

L166. How did you determine your sample size?

L179-98. You can delete everything that is generic landscape metric information. Again, these are just measures and there is a rich literature on them. What needs to be rethought here through a review of the literature on urban bird ecology is an explicit statement of what metrics have been found in the past that are important for describing urban birds. This is a rich literature and you should easily be able to set up with literature support an explicit set of metrics that are important to evaluate.

L197-98. This is an inappropriate statistical approach for use here. Also, there is no clear connection to why this analysis would be done relative to the main research hypotheses or questions.

Table 1. Please cut. You need to go into the urban bird literature and look up what actually is important for describing urban birds.

L201. This methodology is inappropriate. First, it is unclear what you are testing (i.e. what are you dependent and independent variables). Second, you have not evaluated collinearity amongst variables or indicated if the variables follow or were tested for normality. Third, AICc should be used given sample size, not AIC. Fourth, to appropriately use AICc you need to set up an a priori suite of specific models you are going to test, not data mine with forward and backward model selection. Notably, you should consider interactions or non-linear terms. Finally, are you looking at raw richness or abundance or are you correcting these for detectability?

Results-Discussion. I read the remainder of the manuscript, but have not provided additional comments as the work up to this point has a number of significant flaws the preclude the work being publishable at this point. If the data can be reanalyzed and corrected and then be used to test urban bird hypotheses, I would expect an entirely different set of results to be presented.

Author Response

***Author responses

Reviewer 1

General Comments

Overall I was very unclear what the goal of the research was or how it adds to our knowledge about urban bird ecology. As written the manuscript did not clearly convey whether it was about urban bird ecology, testing basic landscape ecology ideas, or simply an evaluation of where birds are located. The Introduction lacked a clear narrative about the field of study and what is known and where gaps in knowledge are. Moreover, there were no a priori hypotheses. Essentially the field of urban bird ecology was not discussed and the ideas that were are not really needed or novel ideas. In the case of Methods there are a number of shortcomings on experimental design, how bird data were collected, and inappropriate use of statistical methods. Ultimately, what is needed is a thorough read of the urban bird literature to identify gaps and laying out a priori hypotheses that the data you collected can address.

***Thank you for your comments. We have undertaken a thorough rewrite of the article, ensuring that there is more up to date literature, the flow is cleaner so it should not be unclear where our article is coming from and what we are arguing. We do however rebut several of the comments regarding the methods and results. These comments appear to come from missing information regrading testing for normality, sampling strategy, etc. However, all of this information was included in the original submission, and we highlight the instances where this information was present in our responses. This means our methods did not have significant shortcomings as suggested, the statistics were not inappropriate and our results can be interpreted robustly, albeit with the stated caveats.

Specific Comments

L10. Here and throughout ms please separate these as urban green spaces and urban blue spaces to be consistent with literature. Also, please avoid use of acronyms and simply write out words.

***There is actually quite a lot of inconsistency in the literature. Several studies separate the terms that you are clearly familiar with, but other researchers keep them together (see the following references). We feel, and argue in the discussion, that there is a need to both separate and keep the terms consistent, because it is the configuration of habitats not the sole presence of individual habitats that is fundamentally important for biodiversity. Finally, it is common practice to use acronyms in our discipline and the journal guidelines are quite clear on the matter. Please see them here: https://www.mdpi.com/journal/geographies/instructions

Kronenberg, J., Haase, A., Łaszkiewicz, E., Antal, A., Baravikova, A., Biernacka, M., Dushkova, D., Filčak, R., Haase, D., Ignatieva, M. and Khmara, Y., 2020. Environmental justice in the context of urban green space availability, accessibility, and attractiveness in postsocialist cities. Cities, 106, p.102862.

de Manuel, B.F., Méndez-Fernández, L., Peña, L. and Ametzaga-Arregi, I., 2021. A new indicator of the effectiveness of urban green infrastructure based on ecosystem services assessment. Basic and Applied Ecology, 53, pp.12-25.

Russo, A. and Cirella, G.T., 2018. Modern compact cities: how much greenery do we need?. International journal of environmental research and public health, 15(10), p.2180.

L35. This is a bit confusing as we generally think of these types of spaces falling within urban ecosystems. While ecosystem is not scale explicit, I would argue for many of the green spaces that these are not in fact ecosystems. Also, this is a run-on sentence. I would thus suggest revising this section.

***This section has been reworded to reduce any confusion in the semantics whether these are ecosystems, habitats, spaces, etc. We have also provided a key citation that states we should be considering cities/urban areas as ecosystems.

L37. Change environs to ecosystems. We consider these to be urban ecosystems.

***This has now been deleted with the rewrite.

L38-43. I would suggest deleting these sentences as your ms is focused on urban ecosystems.

***These sentences were inherently linked with urban ecosystems, so we are somewhat confused by this comment. Regardless, we have edited the section to aid clarity:

“This deterioration is directly linked to human activity, with urban areas ranked as one of the primary drivers of this loss, and the driver with the largest global impact [8]. Of note within this report is the implementation of nature-based solutions, including increasing ecological connectivity within urban areas [8]”

L45. This is also important to city managers, conservation practitioners, etc. I would argue it is actually of more interest and importance to conservation and management and human wellbeing than it is to biogeography and landscape ecology.

***We have now deleted this as its not central to the research

L46. Can delete the definition of connectivity. Readers should understand this concept and if not, then maybe the article is better suited for a standard ecology journal.

***This has now been deleted, but we note that several papers in the discipline do indeed provide such a definition.

L52-72. Again, I would suggest cutting most of these items. Don’t need to review what landscape metrics are and the issue of scale at which these are run has been well discussed in the landscape ecology and urban ecology literature. I think you can greatly simplify such points to be the type of metric used matters as does the scale of the analysis.

***These paragraphs have been shortened and merged.

L73. I’m not clear the importance of this paragraph. If the issue is simply that we have better satellites today, that’s great, but to me this is mainly methodological. That said, you’ve cited papers that indicate we know well what can be included in green space assessments and we have good tools to find them, even small parks, etc.

***The importance of this paragraph lies in justifying the decision to map the UGBS using Sentinel-2, as this has been demonstrated to be best-practice by a number of similar studies, as it allows a level of robustness and repeatability over a large urban area that can’t be achieved by fieldwork alone, and also improves upon the standard land cover maps such as CORINE which don’t indicate small green spaces such as gardens  We have now added an extra sentence to the start of this section to provide more context.

L80. Actually this isn’t really the case. The paper you cite is 13 years old and the reviews you cited earlier in the ms are much more recent descriptions of what a green space is. For instance articles in the last five years are not including water in the term. I feel this paragraph is a bit of disservice to where the field is at currently. Ecologists very well know how to differentiate green and blue spaces. Notably, there may some misunderstandings here in that individual researchers define what they are evaluating in a given paper and certainly there are papers that lump terrestrial and aquatic together for certain types of assessments.

***This is a nuanced point and the second half your comment pretty much explains why we are still seeing high impact articles published in the last 12-months that do not separate these terms in their abbreviations. We felt that highlighting the issue should be of importance for the discipline; however, perhaps the novelty of this paragraph was a bit overstated, and as such for ease of flow, we have deleted the paragraph.

Fletcher, D.H., Likongwe, P.J., Chiotha, S.S., Nduwayezu, G., Mallick, D., Md, N.U., Rahman, A., Golovátina-Mora, P., Lotero, L., Bricker, S. and Tsirizeni, M., 2021. Using demand mapping to assess the benefits of urban green and blue space in cities from four continents. Science of the Total Environment, 785, p.147238.

L90. What do you mean lack of consensus on what configuration matters? I don’t think there has really been much work on testing specific real world configurations against null or random configurations. Also, configuration would be species dependent, so I’m at a loss as to what you mean here.

***This has been clarified to state that we were referring to a lack of consensus on the importance of the specific landscape configuration for biodiversity patterns. We are referring to biodiversity not species specific relationships, but the configuration can still be important for overall richness and abundance. We have now included more citations in the text that support this.

L93-96. These questions really were unexpected. First, there is a rich literature on urban bird ecology that has been missed in the Introduction. Second, there is no clear establishment of what hypotheses need to be tested in urban bird ecology here. Third, the location of the study is not important as you should simply be using a location to test hypotheses. That is, you can move the location information to Methods.

***We are not sure from your tone here whether these questions are unexpected in the manuscript or whether they are unexpected from the literature review, as your comments suggest both. With regards to the urban bird literature, we do include several studies on the subject, so we feel that the ‘rich literature’ that has been ‘missed’ is perhaps a bit overstated. One can always add more literature, but without a guide to specific papers, we cannot cover everything. That said, given the edit of the introduction we now have more space and have added a further paragraph on this topic, specifically outlining key research from the last couple of years.

***Secondly, the statement of specific hypotheses to test is an epistemological view of research that could be considered slightly outdated. Particularly given the exploratory nature of the study to compare models across spatial scales. We will not be stating a set of specific hypotheses to test.

***Finally, we completely disagree that location is not important. This journal is ‘Geographies’, meaning that spatial location is of paramount importance to the study. Yes, it is true that research should be generalisable, but that doesn’t mean research happens in a vacuum irrespective of space. Given the additional explanation we have provided on why we are using remote sensing (and in your next comment), it should become extra clear that spatial location is important to the question in-hand.

L116. I’m unclear why you are building a land cover database for Cork rather than simply using an established land cover database? You have not set up any a priori hypotheses and predictions as to why a novel land cover database is needed. Also, your land cover information should include minimum mapping unit or pixel size, projection, and datum used. Finally, why use 2018 data when your field surveys were in 2019?

***As discussed in the paragraph starting in line 73 Sentinel-2 is used to create a detailed and context-specific land cover map for Cork city as this has been demonstrated to be best-practice by a number of similar studies. It improves upon the standard land cover maps such as CORINE which don’t indicate small green spaces such as gardens as their minimum mapping unit is 25ha, and other land cover maps for the Cork region similarly fail to take into account the dedicated classes for this work. The spatial resolution of Sentinel-2 is identified in line 78, and it is not standard practice to report the projection and datum in such instances as it often reduces the accessibility of the research.

L144. For point counts are accounting for detectability? If not, then this needs to be done. Also, are you using a fixed radius point count and if so what is it. Lastly, how long were your point counts run for?

***We established a rigorous sampling strategy to reduce the impact of detectability on species following literature. This was mentioned in the description of the sampling strategy that you missed (and commented on in L146). We also had discussed the limitation of detectability in the discussion, which you have not commented on. Since the last revision, we have further classified our species into a detectability group, ranking them 1-4. This is included in Table A1. This has been discussed in detail in the discussion, with results caveated against this.

***The length of time point counts were undertaken for was included in the original text (it’s 5-10 minutes). The radius for the point counts was 50m to cover the finest spatial resolution, we have included this in the text now.

L146. What are your target group of birds? Migrants? Breeders? I have no clear idea where in the city you are surveying, what you are surveying for, and how this answers novel or important hypotheses in urban bird ecology.

***This was included in the text. The following paragraph details extensively the target species, the sampling strategy, and where the survey locations were.

L146. Did you only count a location once? Using only one observation in one season is not a valid approach to enumerate birds.

***We appreciate this comment and acknowledge it. However, given the resources and time available for this study, it was only feasible to survey locations once. We have been upfront about this and had outlined the subsequent limitations to our research resulting from this in the discussion, and while a limitation does not compromise the research. We now further outline in our discussion a number of other peer-review studies that have only used one count, and emphasize the exploratory nature of our study.

L148. How are you avoiding double counting?

***This is now included with the relevant citation. Our assumption was this is an implicit feature of the point count method.

L153. You need to describe an experimental design first and then indicate how you surveyed locations within the design.

***This has been completed in detail in the original submission. Please review all the material including SI that was originally submitted. In line with reviewer 2’s comments, we have now moved one map to the main document.

L160. Why? You are working on birds and these don’t necessarily make sense as boundaries.

***This is also explained the original text:

“These classes were sampled throughout Cork City, and sub regions were integrated into the study to stratify the sampling of data. The city was divided up into different zones (sub regions) based on constituencies [40] to allow for a complete spatial coverage of the study area and reduce spatial autocorrelation (Supplementary Information 1). The larger zones were further subdivided again using major road lines or natural features such as rivers and streams as boundaries. The zone borders were modified so each habitat class could be sampled in each of the zones using a stratified random technique, with the exception of the mudflats class as this habitat only occurred in the zones found in the eastern and south-eastern part of the city.”

L166. How did you determine your sample size?

***Based on the need for a suitable n and the size of the city. It’s quite unusual to see a justification for sample size in a paper when it’s not at the lower end.

L179-98. You can delete everything that is generic landscape metric information. Again, these are just measures and there is a rich literature on them. What needs to be rethought here through a review of the literature on urban bird ecology is an explicit statement of what metrics have been found in the past that are important for describing urban birds. This is a rich literature and you should easily be able to set up with literature support an explicit set of metrics that are important to evaluate.

***Combined with R2’s comments who states the opposite and supports the use of the table, we have reduced this text and incorporated it into the table. While there may be a rich literature on these metrics, we simply cannot use them as variables in our models without first explaining what they are in the paper. Not all readers will be familiar with them, and to assume they are would be poor practice. Moreover we have used the literature and our experience to establish a set of metrics that are important to evaluate. These ecological justifications are present in this paragraph.

L197-98. This is an inappropriate statistical approach for use here. Also, there is no clear connection to why this analysis would be done relative to the main research hypotheses or questions.

***A comparison of values across scales would warrant a t-test. Therefore, this isn’t an inappropriate statistic, but rather you are requesting more information as to why we have undertaken this. We have now positioned our research more explicitly in terms of scale, with recent research highlighting the need to consider such metrics at different scales, particularly due to inconsistent findings. We wanted to investigate whether the landscape configuration changed as the scale increased, and while this was not a central research question, it is important to provide context and understand our regression models. Therefore, this is not inappropriate, but we have provided an extra sentence to explain why we are doing this.

Table 1. Please cut. You need to go into the urban bird literature and look up what actually is important for describing urban birds.

***Cut, but see point two comments above

L201. This methodology is inappropriate. First, it is unclear what you are testing (i.e. what are you dependent and independent variables). Second, you have not evaluated collinearity amongst variables or indicated if the variables follow or were tested for normality. Third, AICc should be used given sample size, not AIC. Fourth, to appropriately use AICc you need to set up an a priori suite of specific models you are going to test, not data mine with forward and backward model selection. Notably, you should consider interactions or non-linear terms. Finally, are you looking at raw richness or abundance or are you correcting these for detectability?

***We disagree with several of these comments. Much of the information you’ve requested is present in the manuscript in its original format, meaning the methods aren’t in fact inappropriate.

***Firstly, we clearly state what the dependent and independent variables are in the following text:

“Models were created for 50m, 100m and 200m with the dependent variables of richness and abundance. One model was parameterized using only the class metrics (PLAND) for each habitat, and a separate model using both the class and landscape metrics, with the exception of total area (TA), as area was already accounted for with the PLAND metrics in this model.”

***Secondly, we do test for several assumptions, including normality, spatial autocorrelation of the residuals, and homogeneity of variance. This is clearly documented in the following text:

“Using R version 3.6.1 [48], the normality of the richness and abundance data was checked using the “car” package [49]. … Spatial autocorrelation was tested for using Moran’s I, and homogeneity of variance was also tested using Bartlett’s and Levene’s tests.

***We did not report multicollinearity as suggested here. This is primarily due to the exploratory nature of the work investigating the impact of scale on the models. Our premise is that the scale of the analysis will alter the results and subsequent inferences, meaning that to compare across scales we need the same variables in each model. Any correlation among variables is going to change at 50m, 100m, and 200m, meaning in some instances we may have had different input variables at different scales, which would have compounded the results of our research. This is also part of the rationale behind the detailed exploration of the landscape metrics at different scales, so as to provide a comprehensive overview of how these metrics altered at different scales, and what this might mean statistically and ecologically for the models. Regardless, none of the independent variables were highly correlated based on preliminary analysis, and we have now added such an explanation to the discussion where we outline limitations of the study.

***Thirdly, we would argue that the sample size is moderate, not small! Therefore, we feel that AIC is fully appropriate here. There will of course be variation among AIC, AICc and even BIC in all instances, but in our experience these are largely minor, particularly within studies of similar sample sizes to us. However, this point is not independent to your fourth. As you state, with AICc one cannot mine the models and requires an a priori assumption. To establish an a priori assumption for multiple models at multiple spatial extents reduces the accessibility and increases the complexity with which to comparatively evaluate the role of scale on results, which is the key premise. Therefore, the selection of information criterion is certainly more nuanced than stated in your review, and certainly does not invalidate the results of this research.

***Finally, we did incorporate interactions in the model. This is clearly stated in the methods section:

“A generalized linear model (GLM) was implemented using a stepwise procedure using both forward and backward processing with second-order interactions”

Results-Discussion. I read the remainder of the manuscript, but have not provided additional comments as the work up to this point has a number of significant flaws the preclude the work being publishable at this point. If the data can be reanalyzed and corrected and then be used to test urban bird hypotheses, I would expect an entirely different set of results to be presented.

***As outlined above, several of the ‘significant flaws’ have actually been included in the manuscript up until this point. We have rewritten the results and discussion in line with reviewer 2’s comments.

Reviewer 2 Report

I appreciate the focus on UGBS, this is a good choice for your paper and is an interesting addition to the literature. My main concerns center around your validation approach for the classification. I also strongly suggest you revise your results--they are wordy and hard to follow (particularly the discussion of landscape metrics).

More detailed suggestions follow:

1. L61: Connect this paragraph more clearly with the previous. That is, connect multi-scale approach with connectivity or landscape metrics which are discussed in the previous paragraph.
2. L125: Did you use a specific software package for the supervised maximum likelihood classifier? Your references don't seem to be methods papers so additional detail (or supplemental info with code) would be good.
3. L 141: Using the publicly available land cover maps seems to be a suboptimal way of evaluating the performance of the classified product. Using high resolution imagery and visual interpretation or site visits would be better. Additionally, how did you choose the 40 sites? How did you determine this sample size? Best practice is to use random stratified sampling to achieve the desired precision for uncertainty (i.e. the standard error of estimated overall accuracy that you would like to achieve). See: Olofsson, P., Foody, G.M., Herold, M., Stehman, S.V., Woodcock, C.E. and Wulder, M.A., 2014. Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, pp.42-57.
4. L167: I'm surprised that some water classes were not accurate. Why were they not accurate? Sometimes floating vegetation can cause these issues, but in my experience open water is one of the easiest things to classify. I see you address this in your discussion but I suggest you mention briefly your issues here.
5. Additionally, you can always do post processing on your model. You can use additional data sources (e.g. LiDAR, existing layers) to increase accuracy.
6. L173: I'm not sure about your decision to remove S33 and S37 as this is variation you'd expect to see on the landscape. I think if you choose to exclude these sites you either need to more strongly justify your decision or caveat your results as this is excluding valid data.
7. L179: I think you could reduce the word count in this paragraph. Consider putting some of the context into the "Description" column of Table 1 which is very helpful.
8. L217: Please provide confidence intervals for your area percentages.
9. L225: I'm very confused by your water inaccuracy. Could you provide some context about what caused the poor producer's accuracy? It looks like higher order reaches were less well captured.
10. Figure 1. Consider including photos of what each class looks like from the ground (or in high quality aerial/satellite imagery)
11. L241: Some wordiness in this paragraph, consider revising for clarity.
12. Figure 2 & 3: Make title capitalization consistent. Use printer friendly and internally consistent color schemes (e.g. the bright yellow with black outlines of Fig 2 is not thematically similar to the blue and green with no outlines and checked background of Fig 1).
13. L282: This paragraph needs revision--it is wordy and confusing to read. I suggest you delete most of it (everything but the highest level results) and instead put an "Interpretation" column in Table 4.
14. L301: This paragraph similarly needs revision. Additionally, you refer to Table 5 in the first sentence but then go into a discussion of CA, PLAND etc. This should be a separate paragraph (one idea per paragraph is generally a good rule of thumb).
15. Table 6 would be better as a Figure. Use connected points with error bars.
16. L321: Similarly, this paragraph needs revision as it is hard to follow.
17. L336: See comment for L173.
18. L341 and Table 7: Most of the information in this paragraph is redundant with Table 7. The table is very good and I suggest you only discuss the most important results in the text and let Table 7 convey the rest.
19. L345: Given your poor water class accuracy, you may need to qualify any results with this class.
20. Table 7&8: No colors in the table. Don't show values that are not significant.
21. Table 7 & 8: Did you use any corrections for multiple tests? List them or note that you did not in the caption.
22. L370: Same comment for this paragraph--wordy, most information in Table 8.
23. Overall, the discussion needs additional references. Contextualize your research in the broader literature more completely (e.g. many studies have found built class detrimental to bird diversity). Additionally, try to make the discussion more concise.
24. L409: Aha, great. I was waiting for this explanation of the water issues. However this also raises questions. If mixed pixels caused much of the error then the validation data becomes critical. Specifically, how do you know that the land use layer you used was correct? How did you decide which of the competing classes to consider "true"? I really do think you need to revisit your validation approach.
25. Table 9: This is a helpful summary--a comparison with similar sites that are not on the River Lee would further strengthen your case.
26. L502: This may be due to differences in the definition of woodland.
27. L526: I agree and appreciate your addressing survey limitations. I also don't think point counts are a great technique for cities, despite the fact that they are used constantly. You must address a critical issue: using an equal effort approach, sites with greater habitat diversity / low detectability are comparatively under surveyed compared with sites with low habitat diversity/ high detectability.

In general, your supplementary material is great.

S1 maps: Offset the boundaries slightly (draw them inside if you can). Right now it is very confusing to see where the boundaries are for the internal sites (e.g. City Centre).

S1.2 map: Please clean this figure up and make it a figure in the main text.

Author Response

***Author responses

Reviewer 2

I appreciate the focus on UGBS, this is a good choice for your paper and is an interesting addition to the literature. My main concerns center around your validation approach for the classification. I also strongly suggest you revise your results--they are wordy and hard to follow (particularly the discussion of landscape metrics).

***Thank you for your detailed and supportive comments. We have addressed these and feel they have increased the accessibility of the paper. We have in particular explained further the validation approach and have substantially rewritten the entire results and discussion section.

More detailed suggestions follow:

1. L61: Connect this paragraph more clearly with the previous. That is, connect multi-scale approach with connectivity or landscape metrics which are discussed in the previous paragraph.

***We have now shortened both paragraphs and subsequently merged them to aid clarity.

1. L125: Did you use a specific software package for the supervised maximum likelihood classifier? Your references don't seem to be methods papers so additional detail (or supplemental info with code) would be good.

***The classification was undertaken in ENVI5.5. This has now been added.

3. L 141: Using the publicly available land cover maps seems to be a suboptimal way of evaluating the performance of the classified product. Using high resolution imagery and visual interpretation or site visits would be better.

***The publicly available landcover maps are derived from medium and large scale Ordnance Survey products at up to 1:1,000 scale and aerial photography captured between 2013-2018 at 15-25cm resolution. We have now added this information.

4. Additionally, how did you choose the 40 sites? How did you determine this sample size? Best practice is to use random stratified sampling to achieve the desired precision for uncertainty (i.e. the standard error of estimated overall accuracy that you would like to achieve). See: Olofsson, P., Foody, G.M., Herold, M., Stehman, S.V., Woodcock, C.E. and Wulder, M.A., 2014. Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, pp.42-57.

***The total number of sample sites was 150, these were selected by randomly identifying points within the classified regions scaled according to their size (i.e. the mudflat region which had the smallest total area had the lowest number of random points selected within it), this is stratified random, which as suggested is the most suitable for this type of land cover classification. We have now added this, and acknowledge the lack of clarity in our explanation.

5. L167: I'm surprised that some water classes were not accurate. Why were they not accurate? Sometimes floating vegetation can cause these issues, but in my experience open water is one of the easiest things to classify. I see you address this in your discussion but I suggest you mention briefly your issues here.

***A good point. We have now addressed this here. And following on from your future comments, we explicitly continue to address this throughout the ms.

1. Additionally, you can always do post processing on your model. You can use additional data sources (e.g. LiDAR, existing layers) to increase accuracy.

***As suggested in points 4 and 5, some of the issues are now mentioned here, namely the two key reasons for inaccuracy in the water classification, along with the justification for why post-processing was not adopted

7. L173: I'm not sure about your decision to remove S33 and S37 as this is variation you'd expect to see on the landscape. I think if you choose to exclude these sites you either need to more strongly justify your decision or caveat your results as this is excluding valid data.

***A fair point. As these were the only 2 ‘brown’ land covers in the dataset and given the fact that these points skewed the dataset beyond normality, we felt that we could remove them with little ecological impact to the results. These locations were on the edge of the city, so perhaps don’t represent typical urban blue spaces. We have included this extra information here, and also caveat our results in the discussion later.

1. L179: I think you could reduce the word count in this paragraph. Consider putting some of the context into the "Description" column of Table 1 which is very helpful.

***Thanks, a good suggestion. The work count has been reduced and we have added a column to the table providing the ecological justification.

1. L217: Please provide confidence intervals for your area percentages.

*** While we have 150 points overall, for some covers we only have 10 points, meaning we do not hit the threshold of ‘a sufficiently large sample of validation points’. To acknowledge this, we have now provided a caveat within the discussion.

1. L225: I'm very confused by your water inaccuracy. Could you provide some context about what caused the poor producer's accuracy? It looks like higher order reaches were less well captured.

***As mentioned in response to points 4 and 5, some additional material was included in section 3.2, with some additional explanation here – line 245-247

1. Figure 1. Consider including photos of what each class looks like from the ground (or in high quality aerial/satellite imagery)

***While a good idea, now our spatial locations are provided in the main text, such images might cause an overload of information on the reader.

12. L241: Some wordiness in this paragraph, consider revising for clarity.

***This section has been reworded for clarity.

1. Figure 2 & 3: Make title capitalization consistent. Use printer friendly and internally consistent color schemes (e.g. the bright yellow with black outlines of Fig 2 is not thematically similar to the blue and green with no outlines and checked background of Fig 1).

***These figures have now been revised for consistency.

1. L282: This paragraph needs revision--it is wordy and confusing to read. I suggest you delete most of it (everything but the highest level results) and instead put an "Interpretation" column in Table 4.

***Thanks for the suggestion to include an interpretation column in Table 4. We have now included this, and have subsequently rewritten this entire section of the results to make it less wordy and report only the key results. As such, it much shorter and focused.

1. L301: This paragraph similarly needs revision. Additionally, you refer to Table 5 in the first sentence but then go into a discussion of CA, PLAND etc. This should be a separate paragraph (one idea per paragraph is generally a good rule of thumb).

***This paragraph has largely been deleted, with only key results remaining.

1. Table 6 would be better as a Figure. Use connected points with error bars.

***A good suggestion, thanks. We have now changed this to a figure.

1. L321: Similarly, this paragraph needs revision as it is hard to follow.

***We have reworded and rewritten this paragraph.

1. L336: See comment for L173.

***See our previous comment. We have actually removed this information from here, as we felt we may have been duplicating information.

1. L341 and Table 7: Most of the information in this paragraph is redundant with Table 7. The table is very good and I suggest you only discuss the most important results in the text and let Table 7 convey the rest.

***This paragraph has been reduced in text to report only the key findings as suggested.

1. L345: Given your poor water class accuracy, you may need to qualify any results with this class.

***A fair point. We have now mentioned this in the results, and discuss this in more detail within the discussion.

1. Table 7&8: No colors in the table. Don't show values that are not significant.

***For consistency we have removed all colours from the table. As we reported models from forwards and backwards AIC selection, it’s considered best practice to report all variables and coefficients that are returned in the final model, even if not significant.

1. Table 7 & 8: Did you use any corrections for multiple tests? List them or note that you did not in the caption.

***No, we did not use any corrections, and this has now been mentioned in the caption.

1. L370: Same comment for this paragraph--wordy, most information in Table 8.

***Again, this has been reduced.

1. Overall, the discussion needs additional references. Contextualize your research in the broader literature more completely (e.g. many studies have found built class detrimental to bird diversity). Additionally, try to make the discussion more concise.

***We have undertaken a substantial rewrite of the discussion. As such, we have included a lot more literature, have shortened it significantly, and feel the flow is much more effective.

1. L409: Aha, great. I was waiting for this explanation of the water issues. However this also raises questions. If mixed pixels caused much of the error then the validation data becomes critical. Specifically, how do you know that the land use layer you used was correct? How did you decide which of the competing classes to consider "true"? I really do think you need to revisit your validation approach.

***There is indeed the possibility of mixed pixels at the edges of other features, but as these tend to be less linear such mixed pixel effects are restricted to the boundaries of such classes and the internal regions can be interpreted confidently given the combined high producer’s and user’s accuracies

26. Table 9: This is a helpful summary--a comparison with similar sites that are not on the River Lee would further strengthen your case.

***Thanks for the suggestion. We have now added all water classes to this table.

1. L502: This may be due to differences in the definition of woodland.

***We decided to remove this paragraph from the discussion as we felt we needed to reduce the wordiness, and there were more pertinent findings to report.

1. L526: I agree and appreciate your addressing survey limitations. I also don't think point counts are a great technique for cities, despite the fact that they are used constantly. You must address a critical issue: using an equal effort approach, sites with greater habitat diversity / low detectability are comparatively under surveyed compared with sites with low habitat diversity/ high detectability.

***Another good point. We have now expanded our discussion on detectability, provided ‘detectability groupings’ for our species in a new table A1, along with frequency of sightings, and discuss several points related to this caveat.

In general, your supplementary material is great.

***Thanks!

S1 maps: Offset the boundaries slightly (draw them inside if you can). Right now it is very confusing to see where the boundaries are for the internal sites (e.g. City Centre).

***We have now redrawn this figure

S1.2 map: Please clean this figure up and make it a figure in the main text.

***We have cleaned this figure up and incorporated it into the main text

Reviewer 3 Report

The manuscript explore the relationship between landscape configuration at 3 different spacial scales and bird diversity in Cork. They found that green space increased species richness and abundance, while blue space was negatively related. Moreover, the impact of scale affected how blue space was viewed as a connective network within the city. I found that the manuscript is well written, the objetive and results are well described and can help stakeholders planning a better connection for blue and green infrastructures. Besides the caveats discussed in the manuscript I found it valuable and I recommend its publication. Thus, I only have minor suggestions to give which are the following:

L111 Please add here the year that research was undertaken (i.e., 2018).

L112 Any refs showing the new boundaries of the city?

Table 2 and 3. I think these two tables could be put together and present only one.

L249-250 Please add frequency values after for each species.

Figure 4 Please sort the X axis in descending order.

Author Response

The manuscript explore the relationship between landscape configuration at 3 different spacial scales and bird diversity in Cork. They found that green space increased species richness and abundance, while blue space was negatively related. Moreover, the impact of scale affected how blue space was viewed as a connective network within the city. I found that the manuscript is well written, the objetive and results are well described and can help stakeholders planning a better connection for blue and green infrastructures. Besides the caveats discussed in the manuscript I found it valuable and I recommend its publication. Thus, I only have minor suggestions to give which are the following:

***Thank you for your review and constructive comments.

L111 Please add here the year that research was undertaken (i.e., 2018).

***We have now provided specifications of this

L112 Any refs showing the new boundaries of the city?

***We have added a map to SI1

Table 2 and 3. I think these two tables could be put together and present only one.

***Thanks for your suggestion. As per R2’s comments, a new table 3 is provided that displays the traditional error matrix with the user/producer accuracies, it was decided to keep table 2 separate as it considers the distribution of classes across the city and is not a measure of accuracy

L249-250 Please add frequency values after for each species.

***We have re-added these values. These were present in the original submission.

Figure 4 Please sort the X axis in descending order.

***We have now reordered the X axis to descending order

Round 2

Reviewer 1 Report

Overall I was very frustrated with the responses provided in my original review. I am sorry if they felt overwhelming or missed minor items that you felt were not read closely, to which I apologize. However, the arguments against providing hypotheses, illustrating well what we don't know about urban green spaces relative to birds, and the statistical approach were quite frankly incorrect. We have not changed how the scientific method is used or why it is important, we do know a lot about urban green spaces in relation to birds, and what aspects need testing in relation to landscape attributes. Furthermore, to really address most urban bird ecology questions we need multiple samples of birds over time/season and measures beyond richness and abundance. The weakness of the manuscript ultimately are due to a lack of data over multiple points in time (we need more than single point counts in time), lack of using a hypothesis driven approach (which I should note is required by the journal guidelines), lack of fully accounting for what landscape metrics are most informative and needed for urban birds (note here that Riitters and others have demonstrated only a handful of metrics are even important and Flather and Sauer have demonstrated that each metric used should be justified a priori as to why it is relevant to birds), and following the analytical approaches currently being used in urban bird ecology.

Author Response

***We acknowledge your frustration. We have commented on many of the above points in our original response, these have not changed, and we continue to feel this is a difference in viewpoint, not validity of results. We acknowledge the lack of repeatability but have included detailed caveats in the discussion on this from the original submission. We have now incorporated more statistics to control for this.

Reviewer 2 Report

First, your edits have greatly improved the clarity of the Introduction and Methods and have helped significantly the interpretability for someone not involved in the project. Thank you for the significant effort in doing so, and know your paper benefited greatly from it.

I reviewed the information in the cover letter and appreciate the author's standpoints and explanations. That said, I do think the other reviewer had some valid concerns that overlap my own with regard to bird surveys completed. However, this issue is quite frankly with the larger field of urban bird studies, as 50 m point counts with one visit is a common approach, though I would argue not a robust one. I believe the field will need to face a larger question of what methods adopted from wildlands studies are effective in urban spaces, but that is not your fight.

Overall, the Results and Discussion are still the weakest parts of the paper. While much improved, the authors should do another pass to make sure their meaning is clear to readers. Some of the language lacks precision which leads to confusion. For example, "the scale widens" or "the UGBS product." In the Discussion, while some valuable points have been added, the overall organization and flow of the argument has not been improved sufficiently. Other issues include long sentences with many clauses and vague antecedents. The introduction is well written, with good flow, sentence length, and understandability and I hope you can apply this style more to the Results and Discussion.

Introduction

• Generally, I feel this sets up your research much better.
• L54-56: I understand what you are saying, but I suggest rewording for clarity.
• L63: "the" needed before southern
• L74: Be explicit about an example of a land cover source (e.g. NLCD in the US is a 30m resolution which is fine spatial resolution but it is a coarse thematic resolution).

Methods

• Table 1: should be "Full" not "Fully". For COHESION it says "Range of values not yet evaluated" which is confusing. Otherwise, this table is valuable.

Results

• Table 3: I suggest you include a more traditional table style confusion matrix so that readers can see what classes are being confused. You can probably include Table 2 in this.
• Section 3.2: Check journal standards re capitalizing bird common names.
• L251-253: This statement on detectability appears to confound abundance/frequency with detectability. Further, the impact of habitat complexity on detectability includes both the visual landscape (e.g. blocking branches) as well as the auditory landscape (e.g. loud noises, consistent noises).
• L279-286: This paragraph is still difficult to understand, particularly the comments about connective metrics differing values.
• Table 4: I've read this before but it took me a long time to understand this table, as I expected it to be a summary of the metrics at the different plot locations (as opposed to the city).
• Figure 5 is improved.
• L302-304: This sentence imparts no information to me as I am not sure what you are trying to say.
• L305: Be clear what you mean by "the scale widens". Are you zooming out or in? I can't tell from the wording.
• Table 5 states "Yellow is 0.05 and blue is 0.1 significance" but the colors are not there anymore. You need to redo this with the ** or whichever marker you are using. Please also indicate the 0.01 and/or 0.001 levels. I also highly suggest you include an interpretation sentence in the figure caption (e.g. an intercept of 14 species richness means we expect to see Z, the significant built slope of -0.08 means that for every X additional built we expect Y fewer species)
• Table 6: same as above.

Discussion

• L369: It's not so much a UGBS product as a land cover classification map of Cork City. Later you refer to this as the UGBS map. Please be clear and consistent.
• New text added at L370 is confusing because it is wordy, also check verb tense.
• L378: This is true, but this effect is seen when trying to map roads with overhanging vegetation. Including leaf on and leaf off data can help with this, as can making sure that imagery is on nadir.
• L375-L383 The new material added helps to contextualize, however all the discussion of water and issues there should be in one place, either its own paragraph or a paragraph describing classification issues.
• L380 Any cities in particular you compared with? Relevant citations?
• L380-L388: This is still confusing wording.
• L385: What is meant by "higher reflecting classes"? Do you mean higher reflectance? Or that the class is physically higher?
• L388: is cross-pixel different than mixed-pixel? I've never heard the term before so this is confusing.
• L394: at some point you probably want to consider/discuss how your issues with edge mixed-pixels likely influenced your landscape metrics. That is, errors are higher at the edge of two classes (and particularly high at the boundary of specific classes). How does this impact the landscape metric calculations and their expected errors?
• L398-399: I am not sure this is true.
• L400: Similarly, I'm not sure what you argue in this paragraph is true. Why would the edge effect change at these two scales? Do you have evidence that it does? Instead, consider that a parameter should not be considered on its own, but rather in the context of the whole model. The Intercept for 100m is positive, while the intercept for 200m is negative. Consider using envfit.
• L420-425: Increase and decrease compared with what?
• L449: Sentence starting "One of the main" should be a new paragraph.
• L464: Adding a paragraph on caveats is good. Consider splitting it into multiple paragraphs with one topic per paragraph.
• L488: This sounds like future research and should go in the conclusions.
• The discussion doesn't talk about 'invisible' green space very much.
• Both Appendices and Supplemental information is unusual--but up to the journal. I suggest you refer to this as "Appendix Table A1" the first time as I had a hard time finding it.

Author Response

***Author Response

First, your edits have greatly improved the clarity of the Introduction and Methods and have helped significantly the interpretability for someone not involved in the project. Thank you for the significant effort in doing so, and know your paper benefited greatly from it.

***Likewise, thank you so much for your thorough comments and reviews, we greatly appreciate them.

I reviewed the information in the cover letter and appreciate the author's standpoints and explanations. That said, I do think the other reviewer had some valid concerns that overlap my own with regard to bird surveys completed. However, this issue is quite frankly with the larger field of urban bird studies, as 50 m point counts with one visit is a common approach, though I would argue not a robust one. I believe the field will need to face a larger question of what methods adopted from wildlands studies are effective in urban spaces, but that is not your fight.

***We agree. If time and resources had permitted, we would have indeed undertaken more surveys. We have now included further analysis to explore the impact of aggregating our results by zones. For example, the richness values of 16 at site 2 and 10 at site 6 (both woody) were averaged to become 13. We completed this for all classes within each zone, averaging both the response variables and the landscape metrics. We then re-ran our statistical models. Across both class and landscape metrics, we observed slight differences in the variables returned by the stepAIC process as would be expected when using an information criterion. However, we only reported seven instances where the coefficient inverted, out of all possible combinations of variables (including two-way interactions), which is 36 for the landscape metric model and 10 for the class metric model. We discuss these differences in the discussion, and they do not change our overall conclusions drawn from the original findings.

Overall, the Results and Discussion are still the weakest parts of the paper. While much improved, the authors should do another pass to make sure their meaning is clear to readers. Some of the language lacks precision which leads to confusion. For example, "the scale widens" or "the UGBS product." In the Discussion, while some valuable points have been added, the overall organization and flow of the argument has not been improved sufficiently. Other issues include long sentences with many clauses and vague antecedents. The introduction is well written, with good flow, sentence length, and understandability and I hope you can apply this style more to the Results and Discussion.

***We have undertaken a thorough edit and rewrite of the results and discussion section. Hopefully this reads better now

Introduction

• Generally, I feel this sets up your research much better.

***Thanks

• L54-56: I understand what you are saying, but I suggest rewording for clarity.

***We have reworded this

• L63: "the" needed before southern

***Added

• L74: Be explicit about an example of a land cover source (e.g. NLCD in the US is a 30m resolution which is fine spatial resolution but it is a coarse thematic resolution).

***Thanks for the suggestion. We have used the CORINE land cover dataset to reflect our situation in Europe

Methods

• Table 1: should be "Full" not "Fully". For COHESION it says "Range of values not yet evaluated" which is confusing. Otherwise, this table is valuable.

***Changed fully to full and have provided a more useful description of cohesion

Results

• Table 3: I suggest you include a more traditional table style confusion matrix so that readers can see what classes are being confused. You can probably include Table 2 in this.

***New table 3 provided that displays the traditional error matrix with the user/producer accuracies, it was decided to keep table 2 separate as it considers the distribution of classes across the city and is not a measure of accuracy

• Section 3.2: Check journal standards re capitalizing bird common names.

***We have removed all capitalization of bird names

• L251-253: This statement on detectability appears to confound abundance/frequency with detectability. Further, the impact of habitat complexity on detectability includes both the visual landscape (e.g. blocking branches) as well as the auditory landscape (e.g. loud noises, consistent noises).

***A fair point. We included this statement given the emphasis on detectability at the first round of reviews. Given the new text in the discussion, this is now largely redundant, so we have removed this.

• L279-286: This paragraph is still difficult to understand, particularly the comments about connective metrics differing values.

***We have now rewritten this paragraph as per your suggestions.

• Table 4: I've read this before but it took me a long time to understand this table, as I expected it to be a summary of the metrics at the different plot locations (as opposed to the city).

***We have now explicitly stated this in the table caption

• Figure 5 is improved.

***Thanks

• L302-304: This sentence imparts no information to me as I am not sure what you are trying to say.

***We have now removed this sentence. It’s not necessary to the central argument.

• L305: Be clear what you mean by "the scale widens". Are you zooming out or in? I can't tell from the wording.

***We have now changed this to coarsens.

• Table 5 states "Yellow is 0.05 and blue is 0.1 significance" but the colors are not there anymore. You need to redo this with the ** or whichever marker you are using. Please also indicate the 0.01 and/or 0.001 levels. I also highly suggest you include an interpretation sentence in the figure caption (e.g. an intercept of 14 species richness means we expect to see Z, the significant built slope of -0.08 means that for every X additional built we expect Y fewer species)

***Thanks for spotting. We have now updated the table caption specifically. We have not provided an explanation of how to interpret regression models in the caption, as this isn’t usually considered standard practice.

• Table 6: same as above.

***Same as above.

Discussion

• L369: It's not so much a UGBS product as a land cover classification map of Cork City. Later you refer to this as the UGBS map. Please be clear and consistent.

***We have now specifically stated this as classification. And we have changed over instances of this in the paper.

• New text added at L370 is confusing because it is wordy, also check verb tense.

***We have split this sentence and rewritten it to reduce wordiness.

• L378: This is true, but this effect is seen when trying to map roads with overhanging vegetation. Including leaf on and leaf off data can help with this, as can making sure that imagery is on nadir.

***We acknowledge this, and images were used from across the year however there were no cloud-free images available in November, December, January (as noted in line 120) which would be the primary leaf-off months

• L375-L383 The new material added helps to contextualize, however all the discussion of water and issues there should be in one place, either its own paragraph or a paragraph describing classification issues.

***This should now be resolved with our rewrite.

• L380 Any cities in particular you compared with? Relevant citations?

***We have now drawn on citations here.

• L380-L388: This is still confusing wording.

***This has now been changed.

• L385: What is meant by "higher reflecting classes"? Do you mean higher reflectance? Or that the class is physically higher?

***This should now be resolved with our rewrite

• L388: is cross-pixel different than mixed-pixel? I've never heard the term before so this is confusing.

***This should now be resolved with our rewrite

• L394: at some point you probably want to consider/discuss how your issues with edge mixed-pixels likely influenced your landscape metrics. That is, errors are higher at the edge of two classes (and particularly high at the boundary of specific classes). How does this impact the landscape metric calculations and their expected errors?

***A good suggestion. We have added an avenue for future research on L448

• L398-399: I am not sure this is true.

***A fair point. We have removed this from the text during the rewrite.

• L400: Similarly, I'm not sure what you argue in this paragraph is true. Why would the edge effect change at these two scales? Do you have evidence that it does? Instead, consider that a parameter should not be considered on its own, but rather in the context of the whole model. The Intercept for 100m is positive, while the intercept for 200m is negative. Consider using envfit.

***We have now rewritten this section which should clarify the results. We would prefer to avoid introducing ordination plots into our analysis, particularly given comments that we need to streamline the results section.

• L420-425: Increase and decrease compared with what?

***This has now been clarified.

• L449: Sentence starting "One of the main" should be a new paragraph.

***This is now a new paragraph

• L464: Adding a paragraph on caveats is good. Consider splitting it into multiple paragraphs with one topic per paragraph.

***We have added various information to this paragraph, and we have split this paragraph. This is where we now introduce the results of our aggregated analysis to demonstrate that our results are robust despite the caveats.

• L488: This sounds like future research and should go in the conclusions.

***Our view is this should stay in the discussion as it is new material and not directly a finding of the paper, rather a suggestion for future research.

• The discussion doesn't talk about 'invisible' green space very much.

***We have now added in more discussion on this

• Both Appendices and Supplemental information is unusual--but up to the journal. I suggest you refer to this as "Appendix Table A1" the first time as I had a hard time finding it.

***A fair enough comment. We’ve moved this to SI

***Thanks again for your comments, we feel they have greatly improved the manuscript.