This study aims to quantify the degree of mismatch between bird species sensitivity and protected habitat intactness across the Americas using ebird abundance data. The authors find that the assemblages most sensitive to the human footprint are located in tropical ecoregions and Central America, where intact habitat protections are also low. Furthermore, they find that the majority of high-sensitivity species are inadequately covered by intact protected habitat.

Overall, I think this study is valuable, well conducted, and reaches an unsurprising but important conclusion about the current effectiveness of species protections. I have only two substantial comments related to the analysis.

First, I am interested in how the uncertainty from each modelling step propagates through the analysis and the impact this has on the results. In particular, the model relating traits to sensitivity doesn’t explain much of the variation and was initially biased (before correction). How does uncertainty in this model affect the results, particularly given that the species used to parameterise this model (data-rich) might not be representative of the species for which the values are imputed (data-poor)? Also, why was the model biased and does this suggest that the model is not adequately describing the underlying data? It would be interesting to explore these aspects more fully, and perhaps put more details in the main document to increase clarity about this important step. In addition, uncertainty in the abundance model might also be substantial, and I would be keen to see how robust these results are to the uncertainty in both of these parts of the analysis.

Second, for the species-level results, it was unclear to me how the coverage of intact protected habitat was calculated across the species breeding range. Specifically, was the intact habitat for a species aligned with the species-specific habitat requirements? There could be more intact habitat across a species extent of occurrence than is usable by a given species.

My specific comments are given below and I hope they are helpful.

Specific comments

Page 3
L13. “which are likely to drive”
L15. What do you mean here by intended management?

Page 4
L8-11. I presume that you don’t have sufficient abundance data for these species? I think you should say that here explicitly.

Page 5
L7. Wide ecological variation is a bit vague. Can you be more specific?
L38. What’s recent? Can you be specific?
L40. These are large ranges and will result in large differences in effort between species lists. I presume that is controlled for in the analysis? I would state that these are controlled for in the models here so as to avoid undue concern.

Page 6
L16. How many of these traits are known versus inferred from closely related species? Do the traits for rare species tend to be inferred from more common species and could this bias the imputation equation?
L40. This is slightly confusing here because the Venter reference is from after the 2013 date. Can you be clearer about the origins of these maps?
L52. “Used” rather the “considered”. Is there much sensitivity to this buffer size?

Page 7
L39. Is this arbitrary or chosen for a particular reason? I think you should explain in more detail why this last filtering step was taken. I can guess the reasons, but it would be nice to briefly explain your rationale.
P45. Can you use a different terminology here for “link” so as not to confuse the link function in the GAM?
P46. Why 6? Could you use a more objective criteria?

Page 8
L5. Is there some way you could incorporate into these estimates the variation caused by the other variables rather than simply fixing them to their median value? I would be interested to know how much uncertainty there were in these abundance = footprint relationships (particularly across the range of elevation and productivity).
L5. Weighted by what?
L21. How good was this model in terms of explained variance? Were the imputed sensitivities generally quite precise or was there a substantial amount of uncertainty around this estimates?
L49. Could you do some sort of bootstrap simulation where you account for the variation in the sensitivity estimates themselves and in the imputed values simultaneously? I just wonder how these uncertainties propagate through the analysis to affect the results. It might have little effect, but it would be interesting to know how sensitive these assessments are to uncertainty at each step.

Page 9
L31. Were the intact protected habitats matched to the species-specific habitat requirements prior to summing the area? There could be lots of intact habitat that is entirely the wrong habitat for a given species. The species range maps define the extent of occurrence not area of occupancy and just by summing across the species range without removing areas with the incorrect habitat conditions you might overestimate the amount of intact habitat available to the species.

Page 11
L9. My guess would be that the sensitivity for data-poor species is likely to be underestimated here because your model used to impute these values was built on species that are not data-poor (to some degree) and might not capture the tails of the distribution that well.

Page 13
L43. This paragraph is a little difficult to follow. I suggest breaking it up into several distinct points.
L52. “have a high proportion of”.

Page 14.
L43. It would be interesting to know by how much they fall short of adequate coverage. Do we just need to increase coverage slightly in a few areas or are the short falls very large and non-overlapping. Perhaps you could add some summary statistics here.

Page 15
L1. Has this change substantially since 2013 due to increased fire frequency/intensity and accelerations in land clearance in some parts of the continent?

Page 16
L36. Is this R2 from your model? If so, this should be discussed more in the methods/results and suggests that this uncertainty should be accommodated in the analysis. I’m not sure you can say that the results are robust because the results don’t differ between all vs data-rich species because the all group is just data-rich + species with sensitivity imputed from information from the data-rich species. I think you need to do some sort of bootstrapping analysis that accounts for the uncertainty in these estimates for data poor species.

David Baker (University of Exeter, 2021)

Dear Authors,

I have really enjoyed reviewing this manuscript. You have done an excellent job with the revisions and think that the paper will make a valuable contribution to discussions around the important problem of how to improve the effectiveness of protected area networks and protect sensitive species. I have a few very minor comments/suggestions (below), but otherwise look forward to seeing the paper published.

Kind regards
David Baker (University of Exeter, June 2021)

Minor comments:
L235 and elsewhere. In the equations, should the ~ symbol be an = ? They have slightly different mathematical meanings.
L335. Should the Western be capitalise? I’m not entirely sure in this context, but worth checking.
L678. I don’t think you need “ones” here.
Fig. 2. I really like this figure, but could you make the axis labels on E and F larger as I struggle to read the text.
Fig. 3. The black outline is a little hard to see. Could you make it wider or perhaps dashed?