Habitat Use of the Hen Harrier (Circus cyaneus) during the Breeding Season in Spain

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript titled “Habitat preferences of the Hen Harrier (Circus cyaneus) during the breeding season in Spain” is a well-structured work that, in my opinion, provides valuable insights into the breeding ecology of Hen Harriers in Spain. Although the overall impression of the work is positive, I have a few points that I believe should be addressed before accepting it for submission.

Due to the inconvenience of receiving the manuscript as a PDF, I will outline the areas I find weak and suggest improvements.

Firstly, the quality of figures 3 and 4 needs attention. It is almost impossible to extract relevant information from these graphs. While zooming in on a PC is easy, once printed, the figures become illegible.

Another minor issue is the absence of Western Marsh Harriers in the discussion. There are several excellent studies on habitat selection/preferences for this species in Iberia and Greece that should be included in the discussion, especially since both species are related and appear to select human-used areas. For example:

• Alves, M., Ferreira, J., Torres, I., Fonseca, C., & Matos, M. (2014). Habitat use and selection of the Marsh Harrier Circus aeruginosus in an agricultural-wetland mosaic. Ardeola, 61: 351–366.
• Bobola, E., Goutner, V., & Liordos, V. (2018). Foraging habitat selection and differentiation among coexisting raptors across an estuarine landscape. Estuarine, Coastal and Shelf Sciences, 213: 108–114.
• Literák, I., Sidiropoulos, L., Škrábal, J., Valkenburg, T., Krejčí, Š., Dostál, M., Navarrete, E., & Vasilakis, D. (2023). Ranging Behaviour and Habitat Selection of Sedentary Western Marsh Harriers (Circus aeruginosus) in the Mediterranean Estuarine Landscape. Waterbirds, 46(1).

Now, addressing more serious issues within the methods section:

To analyze habitat preferences, you use real points and randomly generated fake points within the area depicted by Kernel Density Estimation (KDE) and then analyze the proportion of points in habitats. However, you do not assess the relative availability of the habitat within the depicted area.

What I mean is that the Harriers’ use of mainly non-irrigated arable land could be due to the majority of habitats within the KDE being of that type, and not necessarily because they prefer it—there might simply be no other option. There might be smaller habitats with a relatively higher proportion of points, considering their availability within the KDE. For instance, a habitat taking only 0.1% of the area might have a disproportionately high number of points. You should weight the proportion of points in a given habitat by the relative availability of that habitat (i.e., the percentage it occupies within the KDE). What you have done so far is not habitat preference analysis but rather habitat use.

Lastly, I do not believe that using basic KDE is a suitable method for breeding data with high frequency (1 fix every 15 minutes). This method was developed when GPS data were scarce and limited in the number of fixes obtained. The main problem is that breeding birds will have many overlapping points around the nest, causing an autocorrelation problem and biasing the kernel results. For this purpose, I recommend using Autocorrelated Kernel Density Estimation (AKDE), which accounts for autocorrelation.

Author Response

Revisor 1 Comments

The manuscript titled “Habitat preferences of the Hen Harrier (Circus cyaneus) during the breeding season in Spain” is a well-structured work that, in my opinion, provides valuable insights into the breeding ecology of Hen Harriers in Spain. Although the overall impression of the work is positive, I have a few points that I believe should be addressed before accepting it for submission.

First of all, thank you very much for your feedback and all the comments and suggestions you have provided. They have significantly improved the manuscript. We have incorporated all your suggestions into the Word document using the track changes feature so that you can review them. Additionally, we have addressed each of these points in this letter.

Due to the inconvenience of receiving the manuscript as a PDF, I will outline the areas I find weak and suggest improvements.

Firstly, the quality of figures 3 and 4 needs attention. It is almost impossible to extract relevant information from these graphs. While zooming in on a PC is easy, once printed, the figures become illegible.

Agreed, we have revised both graphs to improve their readability.

Another minor issue is the absence of Western Marsh Harriers in the discussion. There are several excellent studies on habitat selection/preferences for this species in Iberia and Greece that should be included in the discussion, especially since both species are related and appear to select human-used areas. For example:

• Alves, M., Ferreira, J., Torres, I., Fonseca, C., & Matos, M. (2014). Habitat use and selection of the Marsh Harrier Circus aeruginosus in an agricultural-wetland mosaic. Ardeola, 61: 351–366.
• Bobola, E., Goutner, V., & Liordos, V. (2018). Foraging habitat selection and differentiation among coexisting raptors across an estuarine landscape. Estuarine, Coastal and Shelf Sciences, 213: 108–114.
• Literák, I., Sidiropoulos, L., Škrábal, J., Valkenburg, T., Krejčí, Š., Dostál, M., Navarrete, E., & Vasilakis, D. (2023). Ranging Behaviour and Habitat Selection of Sedentary Western Marsh Harriers (Circus aeruginosus) in the Mediterranean Estuarine Landscape. Waterbirds, 46(1).

Thank you for your suggestions, we have added in the discussion section.

Now, addressing more serious issues within the methods section:

To analyze habitat preferences, you use real points and randomly generated fake points within the area depicted by Kernel Density Estimation (KDE) and then analyze the proportion of points in habitats. However, you do not assess the relative availability of the habitat within the depicted area.

What I mean is that the Harriers’ use of mainly non-irrigated arable land could be due to the majority of habitats within the KDE being of that type, and not necessarily because they prefer it—there might simply be no other option. There might be smaller habitats with a relatively higher proportion of points, considering their availability within the KDE. For instance, a habitat taking only 0.1% of the area might have a disproportionately high number of points. You should weight the proportion of points in a given habitat by the relative availability of that habitat (i.e., the percentage it occupies within the KDE). What you have done so far is not habitat preference analysis but rather habitat use.

Agreed, you are correct. We have revised the entire manuscript to address these two concepts and have replaced the term "habitat preference" with "habitat use." Thank you for your observation.

Lastly, I do not believe that using basic KDE is a suitable method for breeding data with high frequency (1 fix every 15 minutes). This method was developed when GPS data were scarce and limited in the number of fixes obtained. The main problem is that breeding birds will have many overlapping points around the nest, causing an autocorrelation problem and biasing the kernel results. For this purpose, I recommend using Autocorrelated Kernel Density Estimation (AKDE), which accounts for autocorrelation.

Regarding the suggestion to use autocorrelated kernels, we believe that in this manuscript, where we focus on identifying habitat use within the general home range during the breeding season, the results would not vary even if we used AKDE. You are correct that AKDE is more accurate for the overall measurement of the home-range perimeter. However, since the locations for each individual would always fall within the same type of habitat, whether using AKDE or KDE, we feel that redoing all the work (given that all the results are based on the 95% home-range area: percentage of habitat use, graphs, Agricultural Use Index, maps, statistical analyses, etc.) would yield little to no variability in the results, and can therefore be omitted.

Additionally, recent studies published using KDE continue to be accepted because the variations in this type of research compared to AKDE methods are not significant. Examples of articles published using KDE include:

• Morollón, S., López‐López, P., & Urios, V. (2024). A new view of territoriality in large eagles: the territory pre‐exists regardless of their occupants. Journal of Zoology.
• García-Macía, J., Álvarez, E., Galán, M., Iglesias-Lebrija, Juan.José.,Gálvez, M., Plana, G., Vallverdú, Nú., Urios, V., (2023) Home range variability and philopatry in Cinereous vultures (Aegypius monachus) breeding in Iberia. Avian Research.
• García-Macía, J., Chaouni, M., Morollón, S., Bustamante, J., López-Ricaurte, L., Martínez- Dalmau, J. Rodríguez-Moreno, B. & Urios, V. (2023). Lesser kestrels of the same colony do not overwinter together. Current Zoology, zoad028 
• Morollón, S., Urios, V., & López-López, P. (2022). Home-Range Size and Space Use of Territorial Bonelli’s Eagles (Aquila fasciata) Tracked by High-Resolution GPS/GSM Telemetry. Diversity, 14(12), 1082.
• García-Macía, J., Vidal-Mateo, J., de la Puente, J., Bermejo, A., & Urios, V. (2022). Spatial ecology of the Red Kite (Milvus milvus) during the breeding period in Spain. Ornis Fennica,

Morollón, S., Pausas, J. G., Urios, V., & López-López, P. (2022). Wildfire response of GPS-tracked Bonelli’s eagles in eastern Spain. International Journal of Wildland Fire, 31(9), 901-908.

Reviewer 2 Report

Comments and Suggestions for Authors

Overall, this is an interesting and potentially useful paper but I feel some aspects of the analyses could be improved or, at least, justified more. My main concern is that Johnson's 3rd order selection looks at selection within the range but the range has already been selected from the wider landscape. This is hinted at (e.g.  lines 42 and 264) but I feel this topic needs more explanation.  Range selection appears to be important given that you identify two range classes. If you can find a copy I strongly recommend Buskirk, S. W., and J. J. Millspaugh. "Defining availability and selecting currencies of use: key steps in modeling resource selection." Resource selection methods and application. Western EcoSystems Technology, Incorporated, Cheyenne, Wyoming, USA (2003): 1-11.

Line 89 - the capture does not avoid stressing the animals, rather it minimises the stress experienced.

Line 115 - does this response variable need some weighting? For example, 100 to 50 is a 50% reduction but so is 2 to 1.  Isn't the 100 to 50 reduction more important? Also, perecentage uses are not independent, for every percentage reduction there has to be a corresponding percentage increase. This becomes important when almost all records are associated with a single habitat type.

Line 138 - isn't this only reliable for a brooding female?

Table 1. What do you know about breeding success? The 1B8D731E and 180225 ranges seem implausibly large for successful breeding attempts, particularly when compared some other females. Does the difference in the 2021 and 2022 range sizes for 200452 reflect a difference in breeding success? Does the range size difference between agricultural and forest harriers reflect differences in breeding outcome?

Line 284 - Geary et al did not consider breeding success so joining it with the Redpath et al citation is misleading.

When describing the results of your glmm it would be useful to know how the explained variation is partitioned between the fixed and random factors. If the fixed factors explain a tiny amount of the variation it wise not to overinterpret the results. See Lüdecke, D. sjPlot: Data Visualization for Statistics in Social Science. R package version 2.8.16, https://CRAN.R-project.org/package=sjPlot 2024 and https://strengejacke.github.io/sjPlot/articles/tab_mixed.html

Would lines 112-115 be better in Section 2.3?

Alan Fielding

Comments on the Quality of English Language

There are some phrases that I would write differently but there is nothing which distracts from the meaning. The only exception is on line 89.

Author Response

Revisor 2 Comments

Overall, this is an interesting and potentially useful paper but I feel some aspects of the analyses could be improved or, at least, justified more.

First, we would like to thank you for all your comments, suggestions, and insights. We have made the revisions in the document using Word's track changes feature so that you can review them more easily and quickly. Thanks to your assistance and all these modifications, the manuscript has improved significantly.

My main concern is that Johnson's 3rd order selection looks at selection within the range but the range has already been selected from the wider landscape. This is hinted at (e.g.  lines 42 and 264) but I feel this topic needs more explanation.  Range selection appears to be important given that you identify two range classes. If you can find a copy I strongly recommend Buskirk, S. W., and J. J. Millspaugh. "Defining availability and selecting currencies of use: key steps in modeling resource selection." Resource selection methods and application. Western EcoSystems Technology, Incorporated, Cheyenne, Wyoming, USA (2003): 1-11.

Thank you for your suggestion. We have modified the text to explain that what we are showing in this manuscript is the use of available habitat within the home range of each individual.

Line 89 - the capture does not avoid stressing the animals, rather it minimises the stress experienced.

You are correct, we have modified this sentence.

Line 115 - does this response variable need some weighting? For example, 100 to 50 is a 50% reduction but so is 2 to 1.  Isn't the 100 to 50 reduction more important? Also, perecentage uses are not independent, for every percentage reduction there has to be a corresponding percentage increase. This becomes important when almost all records are associated with a single habitat type.

In this analysis we are comparing the percentages of actual habitat use (using the locations of harriers within each habitat type in their home-range) versus the sample of potential use (using the same number of random locations in the same home-range). If the percentage of actual locations is greater than the number of random locations, then the habitat is positively used, and the difference is significant in the LMM analysis. The actual values are weighted with the random values, measuring in our study the greater or lesser habitat use of the different types of habitat available in the home-ranges of each individual.

Line 138 - isn't this only reliable for a brooding female?

We believe that this applies to both males and females. Because males also reduce their home-ranges quite a lot even though they have the main functions of territory defence, they also focus their home-ranges on the territorial core, which in this case is the nest, due among other things to the fact that they have to feed the female.

Table 1. What do you know about breeding success? The 1B8D731E and 180225 ranges seem implausibly large for successful breeding attempts, particularly when compared some other females. Does the difference in the 2021 and 2022 range sizes for 200452 reflect a difference in breeding success? Does the range size difference between agricultural and forest harriers reflect differences in breeding outcome?

We fully understand this comment. However, we have an explanation for it. All individuals selected in this study were reproductively successful in the selected breeding seasons. As we explained in the methods both the start day and the end day of the breeding season were selected with drastically reduced and drastically increased kernels respectively of 95% over the whole year. By using the overall kernel of the whole breeding season in this analysis, from the first breeding behaviours with larger ranges, than in the egg and chick incubation season, to the last behaviours when they expand their range again, they are nevertheless still considered as breeding movements.

Therefore, there are some individuals whose movements during the first weeks were much greater than others, especially in the first and last weeks of breeding, coinciding with the reproductive ecology of the species. While others focused much more and much faster in the areas around the nest in all weeks of the breeding period. Here are some examples of both the overall and weekly kernels (Figure 1 and 2) of individual 200452 with some very large weekly kernels and therefore the overall kernel is larger. Or on the contrary, individual 200434 has a very small weekly area during the whole reproductive period (Figures 3 and 4) and therefore its overall kernel is very small as well.

The figures are attached in a Word document.

Line 284 - Geary et al did not consider breeding success so joining it with the Redpath et al citation is misleading.

Thank you, we have clarified this sentence so that it is not understood in this way.

When describing the results of your glmm it would be useful to know how the explained variation is partitioned between the fixed and random factors. If the fixed factors explain a tiny amount of the variation it wise not to overinterpret the results. See Lüdecke, D. sjPlot: Data Visualization for Statistics in Social Science. R package version 2.8.16, https://CRAN.R-project.org/package=sjPlot 2024 and https://strengejacke.github.io/sjPlot/articles/tab_mixed.html

Thank you very much for your suggestion, we have added the conditional and marginal R² values in the text following the results of the LMM model.

Would lines 112-115 be better in Section 2.3?

Thank you, we have changed these explanation to the 2.3 section.

Alan Fielding

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I would like to express my appreciation for the thorough revision of the manuscript. The authors have addressed all of my previous concerns, although figures 3 and 4 would still benefit from larger fonts, as the text is very hard to read at the current resolution and size. Nonetheless, all my concerns have been addressed, and the authors have adequately defended their choice of using the "normal" kernel method. While I agree with their reasoning for choosing this method, I believe we should still encourage the implementation of more novel methods in future work. This will help ensure that newly developed methods are utilized and developed further.

One final thought, although the MS in current form is good, I would encourage the authors to add in the end of discussion either how the results can be implemented or what are the new questions/directions that arise for future research.  Especially, since its the first study to study breeding behaviour of Hen Harriers in Iberia using GPS data. 

In conclusion, I would like to thank the authors for their good work, wish them luck and suggest that the paper be accepted. 

Author Response

Reviewer 1:

I would like to express my appreciation for the thorough revision of the manuscript. The authors have addressed all of my previous concerns, although figures 3 and 4 would still benefit from larger fonts, as the text is very hard to read at the current resolution and size. Nonetheless, all my concerns have been addressed, and the authors have adequately defended their choice of using the "normal" kernel method. While I agree with their reasoning for choosing this method, I believe we should still encourage the implementation of more novel methods in future work. This will help ensure that newly developed methods are utilized and developed further.

Thank you very much for your words. We believe that now the manuscript has a better structure and content. 

You are absolutely right. We are working on it in order to do all our analyses from now on with the best possible and available methods. In this time the evolution of methods has to be taken into account as it is changing very fast.

One final thought, although the MS in current form is good, I would encourage the authors to add in the end of discussion either how the results can be implemented or what are the new questions/directions that arise for future research.  Especially, since its the first study to study breeding behaviour of Hen Harriers in Iberia using GPS data. 

We have added a paragraph at the end of the manuscript with these ideas. Thank you for your suggestion.

In conclusion, I would like to thank the authors for their good work, wish them luck and suggest that the paper be accepted. 

Thank you again for investing your time in improving this manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

line 48 - For this reason, this species can be very sensitive to changes in the habitat where it ...... something is missing at the end of this sentence.

line 52 - and later short migrations through the territory...  I think you need to define what you mean by home range and territory as territory is usually assumed to be a defended area within a home range, i.e. it's a subset of the home range see Odum, E.P.; Kuenzler, E.J.  Measurement of Territory and Home Range Size in Birds Auk 1955, 72(2), 128-137. I do not think you are using territory in this way.

My initial comment about the perecentage uses not being independent, i.e. for every percentage reduction in use of one habitat there has to be a corresponding percentage increase in another, does not seem to have been addressed. Figure 3 shows a decline in the percentage of non-irrigated arable land. As this declines something else has to increase to keep the overall sum at 100% so there is some lack of independence.

I was pleased to see the use of a breakdown of the explained variation but there seem to be errors in this and the results from this do not appear to have been used, rather they are just reported. It is difficult to know what the breakdpwn shows as the text appears to contain an error. If there isn't an error the results of the analysis are very surprising. See the Summary of Mixed Models as HTML Table vignette in the sjPlot R help file (sjPlot::tab_mixed). There doesn't appear to be a citation to the sjPlot package.

lines182 - 184    ...conditional corresponding to the effect of combined fixed and random factors were 0.210. The variance explained by fixed factors (R2marginal) was 0.210. It seems unlikely that the conditional and marginal R-squared values are identical as the marginal R-squared considers only the variance of the fixed effects, while the conditional R-squared takes both the fixed and random effects into account. This would imply no variation explained by the random factors which is extremely unlikely as there are clear differences between individuals and years. I fear that you may be left to conclude that almost all of the differences you have observed are due to individual differences (as I have found in some of my previous analyses).

Author Response

Reviewer 2: 

line 48 - For this reason, this species can be very sensitive to changes in the habitat where it ...... something is missing at the end of this sentence.

You are right, we have added the missing verb.

line 52 - and later short migrations through the territory...  I think you need to define what you mean by home range and territory as territory is usually assumed to be a defended area within a home range, i.e. it's a subset of the home range see Odum, E.P.; Kuenzler, E.J.  Measurement of Territory and Home Range Size in Birds Auk 1955, 72(2), 128-137. I do not think you are using territory in this way.

Thank you for your comment, we have modifieed the sentence.

My initial comment about the perecentage uses not being independent, i.e. for every percentage reduction in use of one habitat there has to be a corresponding percentage increase in another, does not seem to have been addressed. Figure 3 shows a decline in the percentage of non-irrigated arable land. As this declines something else has to increase to keep the overall sum at 100% so there is some lack of independence.

You are right. In figure 3 we are comparing the percentages of utilisation of the different habitats in the different latitudes where the home-ranges of each individual are located. As latitude increases, the percentage of non-irrigated arable land decreases and the percentage of herbaceous and/or shrubland associations increases. We have commented on this in the manuscript.

I was pleased to see the use of a breakdown of the explained variation but there seem to be errors in this and the results from this do not appear to have been used, rather they are just reported. It is difficult to know what the breakdown shows as the text appears to contain an error. If there isn't an error the results of the analysis are very surprising. See the Summary of Mixed Models as HTML Table vignette in the sjPlot R help file (sjPlot::tab_mixed). There doesn't appear to be a citation to the sjPlot package.

lines182 - 184    ...conditional corresponding to the effect of combined fixed and random factors were 0.210. The variance explained by fixed factors (R2marginal) was 0.210. It seems unlikely that the conditional and marginal R-squared values are identical as the marginal R-squared considers only the variance of the fixed effects, while the conditional R-squared takes both the fixed and random effects into account. This would imply no variation explained by the random factors which is extremely unlikely as there are clear differences between individuals and years. I fear that you may be left to conclude that almost all of the differences you have observed are due to individual differences (as I have found in some of my previous analyses).

We believe that the explanation for these statistical results is precisely that the sample of individuals is too low to capture a larger variability, suggesting that the model is capturing all the variance explainable with fixed effects, with no contribution of random effects. In addition, it may also be because the response variable is the difference between random and actual use of each habitat type, and not simply habitat use (where there are obvious differences between individuals breeding in shrublands and those breeding in agricultural areas), so that individual differences are diluted.

Since random effects have little or no contribution to the variance of the response variable, in these cases one of the solutions is to perform a GLM instead of a LMM. Therefore, in order to achieve more parsimonious analyses, we replaced the LMM by a GLM with a Gaussian-type error distribution, which showed very similar results to the previous one. In fact, when we have done the GLM we have obtained the same results in the Table 3 as the previous with the LMM.