Accelerated warming of the Southern Ocean and its impacts on the hydrological cycle and sea ice? refers, as does Curry’s comments in the comments. I suspect we’re now at the going-round-in-circles stage, but it is probably worth one more spin.
Curry begins rather gracelessly:
William, all of these issues were discussed ad nauseum over at WUWT, on three threads. These are certainly valid questions, but not particularly interesting ones IMO, which is why I was not motivated to answer them until repeatedly queried (including email) about them.
The WUWT thread(s) are sprawling and generally far from the point, and (when I last looked) failed to cover the important issues in any meanginful or focussed way. And contrary to the impression that Curry is giving, I certainly wasn’t badgering her by email.
Below, I’ve put in Curry’s comment, which is essentially my original bullet-listed “major errors” together with her replies. Note that the “bullet-list” isn’t in the original post: I only added that in the comments when Curry, well, pretty well wouldn’t read the post and reply; it needed to be condensed to that before she would reply. Or, put another way, there is more in the original post that she hasn’t replied to (there is an entire section “Looking at the paper – models” that she has ignored). And I don’t think her replies below are substantive: instead they are evasive. In a sense this is what you’d expect: she is busy, the paper has been published, she can’t afford to admit to any real flaws in it.
1) you use data from 1950-1978 that is clearly meaningless.
Reread the 2nd para of the introduction. The problem with the data is acknowledged. The data from 1950-1978 is not meaningless. There is data in the region during these periods. Missing data in the two SST data sets is filled in by an EOF analysis. The two different SST analyses give reasonable agreement in the period post 1950 (they diverge sharply prior to 1950). Our analysis of the trend is broadly consistent with other assessments of the temperature trend that are cited in the introduction.
This won’t do. It is no good waffling about the EOF analysis filling in the data; it is absolutely clear from the zonal averge plots that WE provides (which in turn simply reflect what everyone, including Curry, knows full well) that before 1978 the data isn’t usable over large regions.
2) this data contaminates the entire (obs) analysis.
The obs analysis is a minor part of the paper, intended to compare with the model simulations that were the main source of data used in the analysis. The whole issue of filling in missing ocean obs using an EOS analysis is definitely troublesome, particularly prior to 1950. In fact it makes me really queasy about the “unequivocal” confidence of the IPCC. William, let me know if you are prepared to throw out both the baby and the bath water on this one.
This is now attack-as-defence: yes, her analysis may well be junk, but in that case so is the IPCC’s, ha ha. Again, this won’t do. The issue is her paper (though if anyone wants to raise the IPCC temperature records, I’m happy to do so, but not here: it is a red herring). Calling the obs analysis a “minor part” of the paper is an evasion, and notice how she has skipped the essential point: does she agree that the missing data contaminates the entire analysis? We don’t know because she won’t say. I think that Curry is not very familiar with EOF analysis, so genuinely doesn’t know the answer. Which would be fair enough, had she simply answered “don’t know”.
3) the hypothesis that you put forward is not novel.
We cite the Zhang 2007 paper that describes a different mechanism that is not inconsistent with ours, but does not include the atmospheric hydrological cycle. I probably read the Manabe et al. paper back in the 1990’s, but didn’t recall it as we were writing this paper. Did any of you (other than Grumbine) actually read the Manabe paper? There is one statement in the Manabe paper that is relevant: ” the reduction in surface salinity resulting from the increase of freshwater supply at the oceanic surface is mainly responsible for the weaker convective activity in the G integration.” This statement is made in a paragraph discussing the deep ocean convection in the Southern Ocean. Manabe doesn’t discuss the increasing sea ice extent in this context. Grumbine connected the dots in the Manabe et al. paper and came up with generally the same idea we did (we came up with the idea via a different route), and describes it in a half sentence. So, our hypothesis is not put forward per se in the Manabe et al. paper. I occasionally check in at Grumbine’s site, didn’t spot his post on the Antarctic sea ice. Note, the Zhang paper did not cite the Manabe paper either; it just doesn’t say much about the Antarctic sea ice.
Since I’ve crit her above for failing to say don’t-know when appropriate: no, I haven’t read the Manabe paper properly. I think her defence, above, is possible but rather weak, and amounts to half admitting the criticism (They mention increasing snowfall in the context of oceanic deep convection but not in the context of sea ice in a comment lower down pretty well admits it). Eli is less kind. In way, I don’t care too much about this issue, as it has no impact on the correctness of the results: it is just part of the general malaise of carelessness.
[Update: I really should ahve read the paper rather than taking Curry's word for the contents. This is yet another example of her carelessness. As Lazar points out in the comments there is far more. The claim for novelty in LC looks very weak indeed now -W]
4) you could have used an extra decades worth of obs data.
The purpose of the obs data was to demonstrate the consistency of the 20th century climate model simulations with the observations. Data from 2000-2010 would not have helped here, since the AR4 climate model simulations do not extend past 2000.
This is evasion / wrong. Using 2000-2010 would have given an extra decades worth of good observations. The C20C simulations end in 2000, of course, but patching them onto the 21st century simulations is quite reasonable.
5) there is no justification for the EOF analysis.
EOF analysis is basically a filtering technique. You can conduct an analysis with the original data, or with filtered data. The latter can clarify the signal. In this particular paper, the EOF analysis didn’t filter all that much. If the study had been conducted with the original data, it would have been more easily understood by a broader audience. The use of EOFs arguably complicated the analysis, but did not in any way compromise the analysis. Jiping Liu prefers to use EOF analysis; I do not. I think I’ve convinced him not to use the EOFs in future papers unless there is a clear advantage that outweighs the addition of the complexity.
I think what this is saying is that yes, using the EOFs was a mistake, so that is good. I could have used a better word than “justified”, or I could have explained it better. What I meant was that there was no justification in the paper for using EOFs. It is just done, with absolutely no hint as to why it is a good idea. Had they attempted to write down why it was a good idea, they might have realised that actually it wasn’t. Also, I don’t think that “You can conduct an analysis with the original data, or with filtered data” is correct here. EOF was their data-reduction technique. They could, instead, have used take-the-trends as their technique. But either way, you can’t use the original data, because (obviously) there is too much of it. You always need some data reduction. Take-the-trends would have the advantage of not corrupting the rest of the field.