The unprecedented 2015/16 Tasman Sea marine heatwave

The Tasman Sea off southeast Australia exhibited its longest and most intense marine heatwave ever recorded in 2015/16. Here we report on several inter-related aspects of this event: observed characteristics, physical drivers, ecological impacts and the role of climate change. This marine heatwave lasted for 251 days reaching a maximum intensity of 2.9 °C above climatology. The anomalous warming is dominated by anomalous convergence of heat linked to the southward flowing East Australian Current. Ecosystem impacts range from new disease outbreaks in farmed shellfish, mortality of wild molluscs and out-of-range species observations. Global climate models indicate it is very likely to be that the occurrence of an extreme warming event of this duration or intensity in this region is respectively ≥330 times and ≥6.8 times as likely to be due to the influence of anthropogenic climate change. Climate projections indicate that event likelihoods will increase in the future, due to increasing anthropogenic influences.

I also did not understand why there was the need to smooth out some of the data. See the PDF for more details.
In summary, I enjoyed the paper and it only needs minor corrections in order to be accepted for publication (which I hope will happen).

Fernando P. Lima
Reviewer #2 (Remarks to the Author): This paper presents report describing various facets of the 2015/2016 marine heatwave in the Tasman Sea. I think it is worthy of publication in Nature Communications, but first needs some revisions. These revisions may not be too substantial in terms of implementation, but I think they are required.
First, the abstract does not present the paper accurately, I think. From the abstract it seemed like this was a single study, with the ultimate goal of understanding the human influence on the Tasman Sea marine heatwave. But it is rather three studies which do not inform each other: one ("observations") describing the physical nature of the heatwave that occurred; one ("attribution") diagnosing the human influence; and one ("impacts") describing suspected impacts. The "observations" subpaper does not inform the "attribution" subpaper, e.g. by forming criteria for evaluation of whether the climate models are fit-for-purpose (because this heatwave was a case of a moving boundary, biases in that boundary dictate whether the process is represented in the models or not). Neither the "attribution" nor "impacts" subpapers inform any of the other subpapers, and the "impacts" subpaper is not informed by the other two either. Is this a badly designed paper then? It seems like it by the impression given in the abstract (and to some degree the introduction), which presents the paper as a coherent analysis. But I think that is a misrepresentation. This is really instead a (relatively) fast "reporting on/describing various aspects of the heatwave" paper, in the same vein as van der Wiel et alii (2016Wiel et alii ( , 10.5194/hess-2016. Therefore the unifying theme is the physical event itself, consisting of a collation of the standard disciplinary reports on the physical and impacts side of an extreme weather event, and an expansion to include a climate attribution component. The is thus expanding the concept of operational reporting. There are potentially pros and cons to this, but overall I think it is a novel concept/product that both warrants dissemination to a multi-disciplinary audience (as in Nature Communications) and would benefit from resulting discussions. This message finally comes through in the final conclusion section, but by that time I had already critically viewed the paper from the wrong angle (and thus maybe had been overly pedantic in my review below).
Second, there are a number of clarifications that need to be made on aspects of the various analyses. I have highlighted these in the specific comments below.
Third, a number of (to me) odd decisions were made in the approach in the "attribution" subpaper. I have highlighted these in the specific comments below. These particularly probably affect the estimates of the confidence range in the quantitative results, some decreasing and some increasing the range, relative to other more obvious approaches. Either some more obvious decisions should be used or a discussion of the implications of these decisions should be included.
Specific comments: line 22 Some context is needed here. One of the strongest El Nino events ever recorded was taking place at that time, and from the point of few of marine life in the eastern tropical Pacific that was a pretty "long and intense" event.
lines 28-29 Some indication of the range of uncertainty on these numbers is needed. line 29 "and event" -> "and an event" line 29 "Future projections" -> "Projections of the future" lines 38-39 Is "either the longest or most intense ever recorded" since 1982 as well? This is ambiguous.
lines 51-52 "This event impacted... detrimental stressors on coastal fishery and aquaculture industries including abalone, Pacific oysters and Atlantic salmon". The examples sound like the species being fished rather than stressing the fishery.
lines 51-56 Do you have evidence, e.g. references, to back up these assertions? line 106 So there was a bigger one in 1982?
lines 106-108 How much local data is informing the circa 1900 HadISST estimates? Is it interpolation/extrapolation? There is less month-to-month variability, which makes me wonder if it is highly interpolated.
lines [136][137] The "strong southward surface flow" is not clear in Figure 4. The caption indicates that colours indicate actual current speeds (but it looks to me like they indicate current speed anomalies), but there is no directional information. The arrows represent anomalies, so we cannot interpret them in terms of actual currents without knowing the reference current map. line 147 Figure 4 is not much help here, because it shows anomalies, not actual currents.
lines [160][161] I am a bit confused as to what mechanisms are included in "air-sea heat flux" here. Presumably sensible and latent heat mechanisms and longwave radiation are included. But is shortwave radiation? That has bypassed the "air" entirely.
lines 235-237 Why are you mentioning this, seeing as you are using the largest event, not second-largest, to set your threshold (as stated in lines 231-233 and 238-239)?
lines 231-233 I am not sure why you are using this reference from the distant past for estimating the threshold for the attribution calculations. If one is using an absolute (rather than return-period-based) threshold, then the need for a reference period is to remove model-observation bias. As far as the model simulations are concerned, it does not matter which period you choose: there is, by definition (ignoring drift), no bias between the historical and natural historical scenarios. But for the observations it does matter: the quality of the record is much poorer during some periods. It seems to me that you have chosen a period with nearly the poorest quality possible within the 20th century. It took me a little while to match these values to the calculations in the text, because originally they were presented to one significant digit only, whereas here you introduce more precision.
lines 430 and 433 Do shortwave radiation fit in "air-sea heat flux". It is not a flux of heat between the two, but then I do not see where else it would go.
line 471 I think you mean "particularly for the historical experiment", because that has fewer simulations in your table. lines 489-490 If I understood correctly, you are using an absolute threshold, which also means that the frequency of exceedence will be affected by relative bias in the mean ("a"), trend ("bt"), and annual cycle ("T_t^S") components. Did you account for any of these as well? It does not sound like it here. Similarly, for 1982-2005 you are essentially estimating historicalNat in part from CanESM2 and CNRM-CM5, but not for historical. If you are calling these "PDFs", then this sampling represents an unobvious distribution on your prior. There are a number of possible "uniform-like" priors, most of which would involve weighing models by the inverse of their sample size for a given scenario. In any case, this choice of prior needs to be justified.
lines 532-533 Which means your estimated uncertainty range are probably about sqrt(2) larger than the uncertainty around the estimates value? lines 536-537 Um, this is manipulating results. First, if this occurs less than 5% of the time it will not affect your estimate of the 90% confidence interval. Second, if that possibility is within the resolution of your frequency estimates, then that is the result. If we know that a 2015/2016-like event occurred in a historicalNat-like boundary condition in the real world, then we can state that a lower bound of 0/N frequency is impossible, but we cannot automatically infer than an actual frequency of less that 1/N is also impossible. This manuscript discusses an extreme temperature anomaly during 2015-2016 in the Tasman Sea. The paper, as the authors noted is a systematic documentation, which discusses the evolution, causal factors, impacts, and future reoccurrence of the extreme event. The major conclusion of this paper is that the southward advection of warm water was the dominant driver of the extreme event and the likelihoods of this kind of event will increase due to anthropogenic influence. The authors claim that this paper is "an advancement" of previous studies, which focused on different aspects (physical drivers, ecological impacts, and the role of anthropogenic forcing) of similar extreme events occurred in other parts of the world's ocean. I have reservations about this statement, because the scientific methods in this paper are mostly, if not all, based upon these previous studies and piecing together different aspects of a problem doesn't necessarily mean scientific advancement. Moreover, trying to cover every aspect of the problem in a short contribution might instead dilute the core science. With that said, this is a timely report on an increasingly important issue, which should be of interest to a diverse readership. I have some comments below to help the improvement of this manuscript.
Major comments: 1. Perhaps the most exciting scientific aspect of this paper is the investigation of the physical drivers of the extreme event. Looking at Figure 3 to Figure 6, I generally agree with the authors that ocean advection should be the primary driver. However, the key statement in the abstract "the warming was dominated by anomalous net southward advection linked to the East Australian Current" was not well supported. In Figure 6, TH is the total horizontal advective flux that contains advective fluxes from both directions and doesn't directly indicate the role of southward advection of the western boundary current. Figure 4 shows some southward surface geostrophic current anomalies, but doesn't directly show the depth-averaged (0-100m) heat divergence anomaly, which should provide direct evidence to support the statement of southward advection. My suggestions would be to decompose the total advection term in Figure 6 to quantify the contributions in both directions, and/or to plot depth-averaged heat divergence anomaly. 2. The other important discussion of this paper is on the future reoccurrence of extreme temperature anomaly like 2015/2016. This was discussed from page 10-14. However, I found the discussions confusing in that the numbers in the text were not obviously supported by the figures. For example, in the key conclusion "an extreme warming event of this duration was 8.5 times as likely to occur due to anthropogenic climate change; and event of this intensity was 5.9 times as likely", the numbers 8.5 and 5.9 were not discussed/explained in the text. I have listed some other similar issues below. This part would be much improved if the authors can be explicit about the derivation of the numbers. Other comments/corrections: 1. I would always be cautious about using the word "unprecedented". The fact that the magnitude of the anomaly was the largest on record doesn't mean it is unprecedented. 2. Page 6, line 107: "top 10 greatest" was not obvious from the Figure

Response to Reviewers
We would like to thank all reviewers for their careful reading and constructive comments on this manuscript. We believe it is now much stronger after addressing their concerns. The reviewer's comments are listed point-by-point below with the reviewer's original comments in italics and our responses in bulleted roman text.

Response to Reviewer #1
This manuscript by Oliver et al. was a pleasure to read. These authors made a quite comprehensive study of the marine heatwave that occurred in the Tasmanian Sea in [2015][2016]. Each one of the approaches they took is not innovative per se, but combining all those perspectives in a single study is impressive and will certainly become a reference to which other papers will be compared in the future.
The English is good and the tone of the paper is also good. It has a lot of detail which makes it quite long but that is necessary because the strength of this paper would be lost if authors would be forced to cut pieces of the study. And each one of those pieces needs to be properly explained (to assure, for example, the reproducibility of the study), so I think there's no way to make the paper shorter. I am also happy with the statistical analysis that were done. Apart from what is written in the PDF (and in the word document), I think authors should make very clear which reference periods they use for calculating the anomalies. That varies through the paper accordingly to the dataset that is being analysed and obviously, it is not possible to use a single reference period. The reader, however, soon becomes lost and should be helped with more clear statements of the periods and also with explanation for why were those period chosen over any other periods. The reference periods have a strong influence on the resulting trends and so the rationale behind each one must be stated.
 Done.  The following section has been added to the end of the Methods: "Reference periods Note that a number of different reference periods were used throughout this study and these are clarified here. The observational description of the event from the NOAA OI SST data is relative to the 1982-2005 period (from the start of the data to the end of the CMIP5 historical period, to enable comparison). The nearshore temperature logger data anomalies are each relative to the period corresponding to each logger's full time span (see Supplementary Table 1). The CFSv2 anomalies are relative to the full period of data obtained (1 Jan 2012 to 31 May 2016). In the climate change attribution analysis all results are presented relative to the 1881-1910 period (an early period, before significant anthropogenic climate change, that is also well-observed in HadISST)."  In addition, the reference period has now been explicitly stated in Figure captions where this information was previously lacking (e.g. Figs. 3 & 5).
I also did not understand why there was the need to smooth out some of the data. See the PDF for more details.
 The only smoothing undertaken was on the climatology, used to define marine heatwaves. This smoothing was implemented according to the recommendations of Hobday et al. (2016; see point #3 in "Recommendation and conclusion" of that paper). The smoothing was required in order to obtain a smoothly varying climatology, from highly variable daily data, such that the marine heatwave detection algorithm would not lead to erroneous detection of events. In summary, I enjoyed the paper and it only needs minor corrections in order to be accepted for publication (which I hope will happen).

Response to Reviewer #2
This paper presents report describing various facets of the 2015/2016 marine heatwave in the Tasman Sea. I think it is worthy of publication in Nature Communications, but first needs some revisions. These revisions may not be too substantial in terms of implementation, but I think they are required.
First, the abstract does not present the paper accurately, I think. From the abstract it seemed like this was a single study, with the ultimate goal of understanding the human influence on the Tasman Sea marine heatwave. But it is rather three studies which do not inform each other: one ("observations") describing the physical nature of the heatwave that occurred; one ("attribution") diagnosing the human influence; and one ("impacts") describing suspected impacts. The "observations" subpaper does not inform the "attribution" subpaper, e.g. by forming criteria for evaluation of whether the climate models are fit-for-purpose (because this heatwave was a case of a moving boundary, biases in that boundary dictate whether the process is represented in the models or not). Neither the "attribution" nor "impacts" subpapers inform any of the other subpapers, and the "impacts" subpaper is not informed by the other two either. Is this a badly designed paper then? It seems like it by the impression given in the abstract (and to some degree the introduction), which presents the paper as a coherent analysis. But I think that is a misrepresentation. This is really instead a (relatively) fast "reporting on/describing various aspects of the heatwave" paper, in the same vein as van der Wiel et alii (2016Wiel et alii ( , 10.5194/hess-2016. Therefore the unifying theme is the physical event itself, consisting of a collation of the standard disciplinary reports on the physical and impacts side of an extreme weather event, and an expansion to include a climate attribution component. The paper is thus expanding the concept of operational reporting. There are potentially pros and cons to this, but overall I think it is a novel concept/product that both warrants dissemination to a multi-disciplinary audience (as in Nature Communications) and would benefit from resulting discussions. This message finally comes through in the final conclusion section, but by that time I had already critically viewed the paper from the wrong angle (and thus maybe had been overly pedantic in my review below).
 Done.  We acknowledge and, on reflection, agree with the reviewer's comment. We have endeavoured to shift the tone of the abstract and introduction to better reflect the nature of the work: i.e. that it combines the findings from analyses of several factors associated with this extreme marine heatwave event.  We have added the following line to the abstract: "We report on several inter-related aspects of this event: observed characteristics, physical drivers, ecological impacts, and the role of climate change."  The third paragraph of the introduction has been revised as: "This paper discusses the 2015/16 Tasman Sea marine heatwave from observations and ocean models, diagnoses its physical drivers and the role of anthropogenic climate change, and describes the ecological impacts that occurred. By collating the analyses together we can integrate the inter-related consequences of this extreme event from the physical drivers and climate change, and their impacts on marine ecosystems. Thus we illustrate how to characterise marine heat waves regionally, where there is a growing need for a clear and timely analyses for such events."  In the last paragraph of the discussion, we have removed the original statement that the study represents an "advancement over previous studies". We have now revised this statement to read "This is in contrast to previous studies which have separately examined events' physical drivers 3-7 , ecological impacts 7-11 or the role of climate change 9-12 ".  [See also Reviewer #3, Comment #1] Second, there are a number of clarifications that need to be made on aspects of the various analyses. I have highlighted these in the specific comments below.
Third, a number of (to me) odd decisions were made in the approach in the "attribution" subpaper. I have highlighted these in the specific comments below. These particularly probably affect the estimates of the confidence range in the quantitative results, some decreasing and some increasing the range, relative to other more obvious approaches. Either some more obvious decisions should be used or a discussion of the implications of these decisions should be included.

Specific comments:
line 22 Some context is needed here. One of the strongest El Nino events ever recorded was taking place at that time, and from the point of few of marine life in the eastern tropical Pacific that was a pretty "long and intense" event.
 Done.  Re-phrasing to emphasize that it was locally the longest and most intense event.
lines 28-29 Some indication of the range of uncertainty on these numbers is needed.
 Done.  These statements have been presented as the 90% lower confidence bound, rather than the best estimate (median) result.   Done.  It was the largest on record, and records began in 1982. Removed "since 1982" from the text to make the statement clearer.

lines 106-108 How much local data is informing the circa 1900 HadISST estimates? Is it interpolation / extrapolation? There is less month-to-month variability, which makes me wonder if it is highly interpolated.
 The HadISST data is interpolated and therefore provides valid "data" in the presence of low (or no) observational density. An estimate of the presence of absence of underlying data can be obtained from the HadSST3 dataset which consists of monthly spatial means over 5degree grid cells, based on the same in situ observations (Rayner et al. 2003). Averaging HadSST3 over the SEAus region provides a monthly SST time series with missing values when observations are completely lacking. Since 1960 there have been no missing months; and all other decades except the 1850s, 1860s and the 1940s have at least 50% of valid monthly data (see figure below, which has been added to the manuscript as Supplementary Figure 8). It is not the case that variability is lower in the earlier part of the time series, as shown by Supplementary Figure 9 which explicitly shows the variance in moving 30-year windows. We do note however that the 1881-1910 period was the best-observed 30-year period prior to the 1960s (86% valid months over this period). We have therefore modified our method to use this period (instead of 1911-1940, which is less well-observed) to correct the daily climatology due to mean warming since this period (see comment relating to lines "231-233" below). Rayner, N. A.;Parker, D. E.;Horton, E. B.;Folland, C. K.;Alexander, L. V.;Rowell, D. P.;Kent, E. C.;Kaplan, A. (2003) Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century J. Geophys. Res.Vol. 108, No. D14, 4407 10.1029 lines [136][137] The "strong southward surface flow" is not clear in Figure 4. The caption indicates that colours indicate actual current speeds (but it looks to me like they indicate current speed anomalies), but there is no directional information. The arrows represent anomalies, so we cannot interpret them in terms of actual currents without knowing the reference current map.
line 147 Figure 4 is not much help here, because it shows anomalies, not actual currents.
 These three comments relate to Figure 4 which was erroneously captioned as "anomalies" when it is in fact absolute currents. This text has been corrected now. Therefore, the current speed colours do match with the current speed direction arrows indicated in the same figure.
We apologize for this error and the confusion it led to.

lines 190-191 I am a bit confused as to what mechanisms are included in "air-sea heat flux" here. Presumably sensible and latent heat mechanisms and longwave radiation are included. But is shortwave radiation? That has bypassed the "air" entirely.
 Done.  We have applied the oceanography disciplinary terminology for "air-sea heat flux" here, which represents the net contribution due to sensible and latent heat flux as well as shortand long-wave radiation (e.g. Deser et al. 2010). We apologise that this may be ambiguous to readers unfamiliar with oceanographic conventions. This has now been clarified explicitly in the "Upper ocean temperature budget" section of the Methods.  This comment raises two issues: • Why use an early base period at all when using an absolute threshold? We are using a relative threshold. Marine heatwaves are defined using the 90 th percentile threshold (seasonally varying) following the Hobday et al. (2016) definition. Therefore, the choice of base period (and therefore the data from which the climatology and threshold are defined) affects what is and is not detected as a marine heatwave. We are interested in defining a marine heatwave in the modern period but relative to a climate with little anthropogenic influence, and then comparing the likelihood of those events in a naturalonly climate, we must use an early base period. Otherwise we are comparing how likely an event that occurred in today's climate is due to climate change, but today's climate already includes the effect of climate change and it is critical to include this signal in the analysis. • We have chosen a base period from a poorly observed period of time. We investigated the issue of observational density further and found that the 1911-1940 period was poorly observed, relative to the later 20 th century or the decades immediately preceding it. Consequently, we have now shifted our reference period to 1881-1910 (see comment regarding line "106-108" above, including the new Supplementary Figure 8) when the observational density is considerably larger, which should make our results more robust. This time period was relatively well-observed, and therefore can constrain the change in mean warming since then rather well (+0.89 O C from 1881-1910 to 1982-2005), and is also early enough to have little anthropogenic climate change warming evident. This has modified the properties of the events being attributed as well as the results of the FAR analysis, all of which has been updated in the revised manuscript.
lines 242-243 What does this confidence interval represent? Is it the spread across estimates for various models?
 Done.  This confidence interval represents the spread across estimates from various models and ensemble members based on the bootstrap resampling approach, as described in the methods. This has now been clarified in the text on this line.

line 243
The rounding error here is misrepresenting the results. FAR=-0.17 equals a Risk Ratio of 0.85, so following the number of significant digits presented in the FAR value, the 90% confidence on the RR value is consistent with a decrease, i.e. an RR<1.
 Done.  We have increased the precision of the results to at least 2 significant digits, and the risk ratios are consistent with the FAR values at the given precision.

lines 244-245
What sort of bias might be introduced by using different periods for the two scenarios?
 The 1982-2005 period in the historical run has less influence from climate change than the 2006-2020 period in the RCP8.5 run, which is actually what we are interested in. Therefore, the 'bias' will be that 2006-2020in RCP8.5 is 'warmer' than 1982-2005 in 'historical'. However, there is no overlap in simulation period (historical ends in 2005, RCP8.5 starts at that point) and so it is impossible to compare the two scenarios over the same time period.
lines 287-288 So was this a just a coincidence then?
 Done.  We have investigated this further and found that a likely explanation is local adaptation at higher (cooler) latitudes. The following paragraph has been added to the "Ecological impacts" subsection: "Interestingly, abalone mortality occurred at temperatures during the peak intensity of the MHW that were more than 7°C below the thermal maximum of 26.9°C for this species, which is found further north along southeastern Australia 55 , indicating there may be local adaptation by south-east Tasmanian  line 308 How is that deduced?
 This is deduced from available observations and anecdotes, drawing from ecological knowledge and expertise of the co-authors. Nonetheless, we have reworded this text to reflect speculation based on local adaptation results (see response to "lines 287-288" above) rather than speculation about acute vs. chronic stress.
line 316 Do you have evidence to support "there is also a keen interest"?
 The authors have considerable experience engaging with stakeholders in the marine environment including the public, state and federal governments, and aquaculture and fisheries industries. For example, reduced tolerance to post-harvest transport and poor condition of abalone during summer in Tasmania has led to seasonal closures for blacklip abalone (Eastern Tasmania) and greenlip abalone (North-East Tasmania), to minimise postharvest losses (http://dpipwe.tas.gov.au/sea-fishing-aquaculture/commercialfishing/abalone-fishery/abalone-closures). The authors are well in touch with the interests around marine heatwaves in the Tasmanian and Australian contexts, and our expertise has informed this statement.

lines 323
It took me a little while to match these values to the calculations in the text, because originally they were presented to one significant digit only, whereas here you introduce more precision.
 Done.  We have increased the precision of the results to at least 2 significant digits (as per comment above).
lines 430 and 433 Do shortwave radiation fit in "air-sea heat flux". It is not a flux of heat between the two, but then I do not see where else it would go.
 Done.  See comment above on the standard terminology in the discipline of oceanography for "airsea heat flux", and the clarifications added to the text in the Methods.
line 471 I think you mean "particularly for the historical experiment", because that has fewer simulations in your table.
 Our original statement is correct. As stated, the historicalNat experiment was the limiting one in terms of the number of CMIP5 models available with daily SSTs ('tos' variable).
lines 489-490 If I understood correctly, you are using an absolute threshold, which also means that the frequency of exceedence will be affected by relative bias in the mean ("a"), trend ("bt"), and annual cycle ("T_t^S") components. Did you account for any of these as well? It does not sound like it here.
 No, a relative threshold is used based on percentiles. Marine heatwaves (MHWs) are defined as being above the 90 th percentile (seasonally varying) and therefore the offset in the mean (a) is not relevant. The only way in which the mean is relevant is the difference in the mean from the historical to the RCP8.5 scenarios, since the historical climatology is used to detect RCP8.5 MHWs. This also holds for the seasonal cycle (Tt S ). Since the threshold is defined as seasonally-varying, we are not concerned with biases against the observations in the seasonal cycle itself. In other words, when the MHWs are detected, the mean and seasonal cycle are irrelevant -only the non-seasonal variability defines how the threshold sits above the climatological mean. Regarding the trend (bt), in essence this is the primary signal which the climate models are providing so we would not want to correct for any perceived error in it. Effectively, we are using the climate models to provide us with a good estimate of the trend.  Done.  We apologise that the text in lines 235-237 was incorrect, and has now been corrected accordingly (see response to comment regarding "lines 235-237" above).
lines 532-533 Which means your estimated uncertainty range are probably about sqrt(2) larger than the uncertainty around the estimates value?
 We very much appreciate this comment, but unfortunately have been unable to interpret further what this means without further clarification from the reviewer. We would be more than happy to consider and address this with further clarification.
lines 536-537 Um, this is manipulating results. First, if this occurs less than 5% of the time it will not affect your estimate of the 90% confidence interval. Second, if that possibility is within the resolution of your frequency estimates, then that is the result. If we know that a 2015/2016-like event occurred in a historicalNat-like boundary condition in the real world, then we can state that a lower bound of 0/N frequency is impossible, but we cannot automatically infer than an actual frequency of less that 1/N is also impossible.
 Done.  We have no longer removed bootstrap samples with FAR values of 1, and removed these lines from the text. Note that this has modified the results somewhat, and the results have been updated throughout the main text. The most notable change is a dramatic increase in the FAR estimates on MHW duration, since events of this duration were so rare in the historicalNat simulations.  We have also replaced the 90% confidence interval (two-tailed "very likely" statement) with the 10 th percentile as a lower 10% confidence interval (one-tailed "very likely" statement) which is more in line with the existing methods presented in the literature (e.g. Lewis and Karoly, 2013, GRL, 40:3705-3709).
lines 540-542 So actually your historicalNat simulations are not a sample of a natural climate, but rather an extended sample of a 1910-1940-like climate with added variance from the extra bias correction steps required?
 Yes, this is true. Our "natural climate" is not purely a simulated "natural / pre-industrial" climate but rather the anomalies of such a simulate climate relative to its own 1911-1940 climatology. We are not the first study to perform this type of analysis as this has also been done recently, e.g., Lewis and Karoly (2013).

line 750
The current speed (colours) and the current speed anomalies (arrow lengths) look rather similar. Are the colours actually showing anomalies too?
 As indicated above, Figure 4 does not present anomalies but absolute currents -now clarified in the figure caption. This clarification should now make the arrows and colours much easier to interpret.
line 770 These triangles are very hard to see.
 Done.  Triangles made larger and moved to a location which makes them easier to identify.

Supp line 47
Add quotation marks to "Complete".