I enjoyed reading this manuscript. The concept is simple and intuitive: that animals habituate to frequent disturbance. However, the authors note that they are aware of no studies that account for this “frequency dependent tolerance” in large scale predictions of disturbance impacts. I expect that this author team is very familiar with the disturbance literature, and this suggests strongly that this gap is real. It is also important to account for this tolerance when predicting impacts of disturbance, as evidenced by the results. Concepts that are simple, novel, and important are hard to come by, and are a sure recipe for success.

The analysis and presentation are straightforward and concise, and this is appropriate given the message. There are a few instances where the simplicity has gone too far, however. The AIC model selection results and parameter estimates are not even presented in the main body of the MS, and when one consults these results, one sees a dubious parameter estimate for one level of the ‘aircraft type’ covariate (see specific comment below). Flight height is a major consideration in most studies of aircraft disturbance, but is alluded to exactly once in the entire MS, in the description of the previous disturbance map, which indicated use of a 450 m for the estimates. See below for some specific suggestions. Site and disturbance frequency are confounded in the experimental design, but we get little information about whether features of the site might alter birds’ responses to aircraft. Predictions have imprecision and bias, and the comparison of variance in Daily Energy Expenditure with and without Frequency Dependent Tolerance ignores this imprecision and bias.

These comments are minor, however, and could be addressed through straightforward revisions. I have provided suggestions for these and others minor issues below. Overall, this study presents a simple concept that will likely be considered “obvious” 10 years from now. Which, in my opinion, is a mark of great science!

Specific Comments:
The title mentions “human” disturbance. To me, this implies disturbance from individual people (e.g., on foot). Although I understand that the authors wanted to make the results more general, “aircraft disturbance” is a more accurate description, in my opinion.

Abstract (e.g., L34, and elsewhere in the MS) – the authors use “intensity” to mean frequency. To me, as a native English speaker, “intensity” usually implies magnitude. An “intense” noise is a loud noise, not a frequent noise. Intense light. Intense pain. In some situations, intensity could imply frequency, but it might be best to avoid the ambiguity and just say “frequency”, when that’s what you mean.

Intro. The introduction could do a better job of introducing the physiological consequences of disturbance. Predicted DEE is a key feature of the results, and is barely mentioned.

L67: “non-lethal disturbances allow animals to habituate” – clarify whether you mean because the individuals are alive to habituate, or because the population determines that the fear is unwarranted.

L81: “Figure 1 illustrates how IN one location”

L98: In the preceding paragraph, you could do a bit more to incorporate the literature on fear – at least 1 sentence referencing the ecology of fear specifically, and the magnitude of physiological consequences (even in the absence of “responses”).

L100: “aircraft potentially belong to the most disturbing disturbance sources”. I don’t think that this statement can be supported without a reference. Dogs on beaches? Also, there is a highly relevant body of literature on the impacts of aircraft overflights in Arctic environments that is easily missed. This work includes experimental tests of flight height and horizontal distances, as well as studies of physiological consequences of aircraft disturbance. See Miller et al. 1994 and references therein.

Miller, M. W.; Jensen, K. C.; Grant, W. E.; Weller, M. W. 1994. A simulation model of helicopter disturbance of molting Pacific black brant. Ecol. Model. 73:293-309.

L136: Flight height is a major determinant of disturbance, and varies greatly with proximity to airport (and also likely for military vs. civilian airplanes). Do you have any information on flight heights?

L159-166: I find this description of the modelling unnecessarily confusing. Is this 1 model or two? “the number of flights that caused disturbance…cbind” seems unnecessary. Is this just your way of describing how you calculated the probability of disturbance? You then go on to describe covariates, but on L161, imply that there is a single explanatory variable – frequency of overflights. Reword to make it clearer that we are (I think) talking about a single model, with several covariates.

This modelling approach assigns flight frequencies to sites, so site and frequency are completely confounded. Thus, any differences in disturbance arising from “frequency” could actually be site effects. This is unavoidable given the study design, but the implications with respect to potential bias need to be discussed in the discussion. E.g., did sites differ in degree of “safety”, size, visibility, species composition, flock sizes, etc.

L172: you don’t need to explain what AIC is, but do need to explain how you interpreted AIC values. E.g., models within 2 units of top model, 6 units, etc.

L180: I’m happy to see physiological costs here…however, I note that it was not directly mentioned in the abstract, nor clearly mentioned in the intro. A brief statement about the physiological consequences of disturbance might be warranted.

L216-226: Given that AIC is mentioned in the methods, this section should report how the AIC values were used to support these conclusions (unequivocal support for the top model in table S1, based on AIC values)

L224: These R² values do not match those reported in table S1 – 42.6% appears to be the value for the interaction between type and frequency.

L240-252: the section reads more like a conceptual model for the introduction (or methods), rather than results.

L268: what are these values? % DEE?

L266: I think you are correct, but this is potentially an unfair comparison. Per L265 – these are not the actual costs, but rather the predicted costs of disturbance, and the predictions have associated uncertainty and/or bias. The constant probability “model” that was used is that probability at one site (presumably this estimate has small error) is equal to the probability at other sites. The predictions would have small imprecision, but are potentially biased (as shown by this paper). In contrast, the frequency vs probability model that you have developed here probably has greater imprecision, but less bias. Comparing the variance of the point estimates is only half of the story. I think it’s fine to do, but you should acknowledge the issue of imprecision in the estimates.

L295: a brief mention of the time period over which habituation occurred would make this comment more informative.

L316: learning NOT TO react

L324: that BEAR LESS RESEMBLANCE TO birds of prey

L367: “while also the quantity of disturbance response” reword – are you suggesting that it’s important to consider all disturbance sources (i.e., including humans), rather than just aircraft?

L375: “spenD little energy”

L383: “clear reasons exisT for it to be common”

Figure 1. I found figure 1 more confusing than illuminating. Partly because the text description is a bit confusing, but also because the examples don’t really relate to the patterns you’re trying to explain. This paper is really about panel C, not b and d. Also, the panels are graphically unclear. The circles in panel a relate to the lines in the legend – A,B,C. I figured that out eventually. However, “disturbance source frequency” could just be “disturbance frequency”, “disturbance response” should be clarified – this is probability of response (not intensity of response, or something else). The legend should be “Disturbance Source – Type A, B, C”. Etc. I think this figured should be reworked or omitted.

Figure 2. The proximity of airports to some sites strongly suggests differences among sites in flight heights, and lower flights should create more disturbance. The observation of LESS % disturbance at sites with more flights, probably the sites near airports, suggests that this potential bias is not a problem. In fact, birds’ tolerance of even the low flights near airports suggests that even your estimates of the tolerance effect might be underestimates. Conversely, people might erroneously conclude that low flight heights don’t impact birds, based on studies conducted near airports. This could grossly underestimate the impacts of low-level flights in remote areas, for example for military operations (if they are only paying attention to flight height and not frequency). This should be discussed.

Figure 5. Clarify that the dot and dashed line show the reference level of aircraft overflight.

Figure 6. Spell our DEE and HTR. Clarify what is within the dashed line – is this entirely intertidal mudflats? Or is this just the boundary of the world heritage area? Also, clarify what’s land and what’s ocean (already on Fig 2).
(Also sentence beginning on L540 needs to be reworded – not a complete sentence at present. When not accounting for FDT does what? Or are you just saying that these are examples of three scenarios? If so, delete “when”.)

Table S2: The parameter estimates for medium airplane, and medium airplane*frequency are suspiciously large. This must be related to unusual results at a small number of sites with these airplanes? Aha – the problem with “medium plane” is clearly visible in Figure 3 – best to combine with another category, because of sparse data?

I appreciate the authors’ attention to detail with the revisions.

It is helpful to know that flight height data were not available. So, while it might have been ideal to know how aircraft flight height relates to disturbance, this is not possible with the available data. I am therefore also satisfied with the authors’ reluctance to speculate about how differences in heights among locations could have biased the results of this (and other) studies.

In regards to this comment/response: L266: I think you are correct, but this is potentially an unfair comparison. Per L265 – these are not the actual costs, but rather the predicted costs of disturbance, and the predictions have associated uncertainty and/or bias. The constant probability “model” that was used is that probability at one site (presumably this estimate has small error) is equal to the probability at other sites. The predictions would have small imprecision, but are potentially biased (as shown by this paper). In contrast, the frequency vs probability model that you have developed here probably has greater imprecision, but less bias. Comparing the variance of the point estimates is only half of the story. I think it’s fine to do, but you should acknowledge the issue of imprecision in the estimates. Reply: The reviewer suggests that the two estimates have different precision, but the reviewer does not explain why they expect precision to vary. We fail to understand why that would be the case and therefore do not see why it would be an unfair comparison or needs to be mentioned.

…The first estimate is a mean probability of disturbance, and the precision is just the standard error of that mean. The second estimate is a model prediction, with imprecision arising from uncertainty in various parameters in the model. However, more generally, the point I’m making is that precision is an important part of the estimates. The patterns in the point estimates are sufficient for the ecological interpretation. However, if the predictions are so variable that they can’t be distinguished statistically, this alters how valuable they are as a “real world decision tool” (L278). Imagine a highly imprecise model that fit a straight line and suggested that disturbance probability was 50% everywhere. This would lead to the appearance of a “homogenized disturbance impact landscape”, but only because the predictions were poor. I’m not suggesting that this is a major problem with your results – just that it’s worth noting that your results are based on the point estimates, without consideration of precision.

The remainder of my comments were satisfactorily addressed. I think that the small changes made throughout have greatly improved the readability, and I continue to find this to be an interesting and important piece of work.

Specific Comments:

L75- “Such an avoidance-driven redistribution of individuals will also result in populations in areas with more frequent disturbance sources HAVING high tolerance.”

L103: “Flight responses…” replace with “Behavioural responses…” (to avoid confusion with aircraft flights)

L104: “an animal’s…” (apostrophe)
L166/167: “observations in the other studies” – I believe that you mean other “sites” or “locations”

L182: “For each combination of dataset, species, and aircraft” (does dataset here mean “site”? I think so…even though you are treating Rottum as two “datasets”, it might be clearer to just say site)

L186: FrequencY

L186: “dataset identity was include as random” – site?

L205: “We used GPS data from small civil airplane data” (omit second data, and say airplaneS?)

L418: The recommendation for standardization of methods seems to be an important one. Could the authors point to a review paper that presents good methodology? Could they be more specific about recommendations here? Or note that a review on the topic is needed?

Figure 1 – I still struggled with this figure. Clarify in the legend that Type A-C are different SOURCES of disturbance. Rather than a “Type A probability” or “Type A mechanism”. Also, in caption, L544-545, reword to “The panels illustrate this point by depicting an example where observed differences in the probability of disturbance in response to three different sources of disturbance (Type A-C) can be explained by different underlying mechanisms”