|Second review of "Emergent constraints on TCR and ECS from simulated historical warming in CMIP5 and CMIP6 models" by F. Nijsse and co-authors. |
The authors are congratulated on carrying out unexpectedly large revisions of their previous paper with a series of interesting findings. With this new material, however, I have also identified a number of issues that prevents me from recommending publication. I would instead suggest the manuscript is returned for major revisions.
1) Section 2.1 is used to formally argue for using 1975 as a starting year rather than 1970 as used in most other studies. However, based on the description I was unable to understand what is being plotted, but if I am right, I don't think it is relevant. I think what is plotted in panel (b) is the standard deviation of forcing with respect to pre-industrial, given that it rises with time. However, for the purpose of the emergent constraint here, this is completely irrelevant, rather what is interesting as a measure of noise is the standard deviation in the forcing change *after* the given year. Likewise, the signal should be the forcing change afterwards.
2) I remain completely confused as to why the authors apply both a running mean and block averaging. In their reply they state that no data outside the blocks is used. If this is true, then years within the block will have been given different weight in order to have results depend on the running mean length. Also, how is it possible to apply a 20-year running mean (Figure 5b) within a block of only 11 years?
3) In linear statistical model used to estimate TCR an offset is assumed (lines 145-147). As for the case of ECS, this offset should be around zero since a world with zero TCR should experience zero warming. Nevertheless, the offset is very large 0.8 or 0.6 K. What is the meaning of this, and why not assume up front that the offset is close to zero?
4) I found the discussion of the different regression methods somewhat one sided. The authors argue against Orthogonal distance regression, but why would one assume zero error in the post-1975 warming estimates from models? Intuitively I would think that the relative error there should actually be larger than the error in estimating TCR.
5) Perhaps not a very major point, but the paper uses many different uncertainty ranges, 16-83, 16-84, 5-95, or even 90 percent confidence with unspecified range. IPCC's 'likely' statement means greater than 66 percent probability, not the 16-84 percentile range. Sometimes these are taken to be the same, but they are not. Practically all studies with which results are compared provide 5-95 percent confidence intervals, and I would recommend that this range is primarily used. I understand that the authors might also want to compare with IPCC, but then they should also state that the 'likely' range is not the same as that provided, but a lower-bound. IPCC can and will add various types of uncertainty not accounted for here and obtain broader likely ranges than found in individual studies.
6) The authors should explain that it is not obvious how to handle the cases of negative ECS with large dT. By removing these cases the error distribution is inevitably skewed and therefore biased towards higher ECS. This is arguably equally unreasonable as keeping the cases in. A reasonable alternative is to remove an equal number of cases from the high end of the error distribution to alleviate the bias.
7) I was confused over the discussion of the different model fits (lines 241-259), which doesn't seem to lead to any new conclusions. Perhaps I missed something, though, the authors might want to clarify what they learned from the exercise.
Minor or technical issues:
31, This sentence is confusing, to first order the reason the ratio is decreasing is because of deep ocean heat uptake. Pattern effects make this effect stronger.
36, Perhaps specify that AR5 used models and historical warming, rather than 'multiple lines of evidence'.
40, The Gregory et al. paper was published 2019.
76, Perhaps reference to Kiehl (2007) would be appropriate.
80, Which 'last decade'?
105, I guess you also need abrupt4xCO2
140, I would suggest to use broadly used nomenclature 'ocean heat uptake efficacy (representing a pattern effect) and \gamma the deep ocean heat uptake parameter'
143, what is 'JM19'?
155, perhaps refer to Hansen et al. 1985 for unrealised warming?
179-180, These specific models can not be distinguished in the plot.
181, which 'dip'?
185-186, how was the confidence interval calculated?
210-211, It seems to me that the uncertainty increases (surprisingly) little with increasing start year (Fig. 5c). Instead it is the median that increases. Also, there is no reason the uncertainty range should revert to the CMIP6 range, it can be much broader if things are done correctly: Imagine an ensemble with very little spread in TCR, then the uncertainty in the relationship should be very large and the posterior spread of TCR too.
Figure 5c, which 'Model range'?
Figure 6b, greens are hard to distinguish, they look the same in print. Also yellow is a poor choice. There are many other colours to choose from.
279, delete 'tests'
290, I believe this should be 1.9-3.4 K