|I appreciate the revisions that have been made to this paper. This version is much improved over the original paper – and indeed, is closer to the paper that should have been submitted in the first instance. |
I disagree with the implied suggestion that allowances should be made due to time pressure associated with the goal of rapid attribution – indeed, if the rapid analysis of events has now evolved into a quasi-operational activity, then I would argue that results should be published elsewhere than in the peer reviewed scientific literature unless specific cases have something to teach us that was not previously known about the methods and tools that are used or the mechanisms that produced the event in question.
While the paper is improved, I do have some additional comments that are listed below. Many of these comments call on the authors to think a bit more carefully about how they are communicating with readers since they do not necessarily share the vocabulary that the authors use to communicate amongst themselves.
51-52: The manuscript mentions in a few places that “ignition sources and type of vegetation[, which are] factors largely independent of meteorology[,] play an important role”, but surely this is not true. Lightning is certainly a major ignition source, and the type of vegetation in a given location is certainly dependent on the climatology of that region. Better wording would be appropriate.
128: Do you mean similarity in the correlation coefficients (rather than confidence intervals)? The discussion goes on to mention explained variance, which implies that it refers to the square of the correlation coefficient.
169-170: Exactly what is an “annual mean low precipitation” value? This seems a confusing combination of words.
185: I think what is meant is that the iterative maximization of the likelihood function does not converge. It is not the “fit”, per se, that doesn’t converge.
193: What is meant by the statement that precipitation is “positive-definite”? Most readers would think about a matrix when they see this term, and wonder whether they missed something in the description of the methods that involved the creation of a matrix of some kind. A few might think about other notions of positive-definiteness as defined by mathematicians, and I think they would also wonder what is meant. Maybe you just mean that precipitation observations are always non-negative?
196-198: I have two comments on this new sentence.
First, ENSO (and other modes of internal variability that affect the region) are certainly relevant since it would be hard to think that the probability of extreme temperature, precipitation and FWI are all insensitive to the phases of these modes of variation.
Second, what kind of externally forced variability are you talking about here? If the interest is in anthropogenic forcing, which has a largely monotonic response over time, wouldn’t a longer time average that better isolates the signal by removing more of the high frequency variation (mostly internal, but perhaps also due to episodic volcanic forcing) be better?
201: Here and throughout, make it clear that all probabilities are ESTIMATED, and thus subject to uncertainty.
204-212: There is something here that I seem to be missing. For example, if the GEV location parameter is a linear function of T’, then how can it also be an exponential function of T’, as in equation (3)?
251: The responses to my previous comments promised that this shorthand (𝜒2/dof) would be clarified – but that appears not to have been done (I could not find a definition of the statistic that is referred to here in either the main text nor the supplement.
271: How did you determine that the factor was “at least two”?
Note also that there is an important nuance of the communication of probability ratios that the paper seems to be sloppy about. When PR=2, p1 = 2p0, and thus the event is estimated to be 2 times AS likely in the current climate as in the counterfactual climate, or equivalently, 1 times MORE likely. If you say “at least two times more likely”, then my interpretation would have to be that PR ≥ 3. This vagueness of interpretation seems to crop up in several places.
330-332: This convoluted sentence will be very hard for readers to understand. I suggest breaking this up into two or three sentences that explain in bite-size chunks the differences between the analyses of the observations and that of the climate model output (rather than trying to make the readers swallow the entire sandwich at one go).
Also, here and elsewhere, be careful with the word “models”. For example, on line 330, the text reads “the models use as covariate the model GMST”. Evidently the first use of “model(s)” refers to the statistical models used in this study, whereas the second use of “model”, only 5 words later, refers to climate models. That implicit distinction will be clear to some readers, but many other readers will be puzzled.
Figure 3 caption: Describe the grey shaded band that is shown in the two right-hand panels.
Figure 5 caption: Describe the shaded bands that are shown!
343-344: The statement that “all model dispersion and shape parameters lie within the large observational uncertainties” seems a bit of a stretch. There is some overlap in the spread of the dispersion parameters from the climate models with that from the observations for 3 of the 4 climate models, but that overlap doesn’t necessarily mean that for those climate models, one would accept the null hypothesis that the dispersion parameter from the climate model is the same as the dispersion parameter from the observations.
353: Results from ERA5 are confounded with the impact of low-frequency internal variability during the single 41-year realization of the observed climate that ERA5 attempts to reconstruct…
355-356: Here is another example of ambiguity in the interpretation of the probability ratio. Judging from the figures, I think you should be saying seven rather than eight times MORE likely, and for the lower bound, perhaps only 2.5 times MORE likely.
367: This might not be as much of a climate model deficiency as implied here (which is what “underestimation” implies). As very briefly discussed in Section 6, and further elaborated in Section S3, apparently the IOD and SAM did play a role.