A new paper by Karl et al. in Science makes a spirited argument that there really is no “hiatus” or “slowdown” in global surface temperature warming. The paper focuses on some of the more technical details of bias correction in the temperature data record, rather than on the dynamics of the climate. It is from a respected group, and from my limited knowledge of the subject seems well put together. I’ll let my more knowledgeable colleagues comment on the bias correction side of things.
The conclusion is a great hook of course, which is why I guess a paper on some small changes to the temperature record is in Science. I have a few thoughts.
1) The changes to the global mean temperature record are, in the grand scheme of things, really quite small. Of course adding two extra years of data in a world that is warming will, probably, increase the mean trend a little bit, bringing it closer to the expected trend from the CMIP5 models. However, the “hiatus” period of 1998-2012 is also estimated by the team to have warmed a little bit more more than previously thought, mostly in the oceans.
2) It doesn’t take much to remove a hiatus. This is what happens if your hiatus definition relies on oversimplified binary statistical thinking, rather than with reference to a deeper description of the climate processes that are happening, in the context of natural variability. I’m not singling out the NOAA group here, this has happened a lot.
My colleague Chris Roberts points out that the necessary comparison is with the expected warming (and associated distribution) from the models, not from the somewhat arbitrary threshold of “zero” warming. Here’s a comparison from our paper on the future of the hiatus, from earlier this year:
It it also worth noting that, even in this new analysis that the 90% CI for the hiatus period (1998 – 2012) still includes zero, when “extra uncertainty associated with our corrections” is factored in. Check out figure 1, box 1.
3) Hiatus communication can be a proxy war. Climate skeptics have long overemphasised the hiatus – they’re unlikely to miss a good opportunity to claim that climate change isn’t happening, or that the models are wrong. There seems to have been a counter effort to demonstrate that the hiatus never existed in the first place. I stand by our comments from last year – the thing branded a “hiatus” is an interesting feature of a complex system; a useful hook for exploring how to make the models better, and how to talk with people about uncertainty. Does it prove that global warming isn’t happening? No. Was it unexpected? Yes, a bit. Are we able to explain it with our current understanding? Yes, very probably. So let’s do that.
What’s more, the last decade-or-so is interesting, and we can learn from it. There were for example, winds in the tropical Pacific that were unprecedented in the historical record (and model run library) and worth explaining. Energy continues to be accumualted by the planet, and lots of it has been buried in the oceans. The pattern of surface temperature change looks like an unusually long La Nina. The hiatus has lead to good work on the uncertainty in recent anthropogenic and volcanic forcing, and it’s interaction with natural processes; the importance (and relative lack) of observations in the deep ocean; the uncertainty in estimates of natural variability in the Earth system.
4) The best way to ensure that a hiatus ends is to publish a paper pointing out that it could well continue for some while. I guess it was inevitable that one group or other would would get to announce the end of the hiatus, but I wasn’t expecting it until next year, after this year’s warming plays out.
Paper:
Doug McNeall's blog 
[…] Doug MacNeal’s take [link]. […]
[…] Doug MacNeal’s take [link]. […]
Doug, I like your summary. But the last sentence is funny. Karl et al. don’t declare the “end” of the hiatus. They are pointing out there is no evidence that it had a “beginning”!
Good post, balanced on the various pluses and minuses.
I think they’ve overstated their main conclusion though. The extra 2 years of data to 2014 are irrelevant in assessing data biases, and no matter when you define the hiatus period start date (1998 or 2000), SM Table S1 shows that the hiatus period rate is only 2/3 that of the pre-hiatus period rate, starting at 1950.
I take it from the above that you accept the that adjustment of calibrated buoy temperatures to match uncalibrated ship engine room intake temperatures is a valid scientific action.
That is, if x1 is the pre-hiatus decadal rate, assuming 1998 as hiatus start, while x2 assumes 2000 start, then:
# 0.086*(15/62) + (47/62)*x1 = 0.129, thus:
x1 = (0.129 – .028) * 62/47
# 116/106 * .086 * (13/62) + (49/62)*x2 = 0.129, thus:
x2 = (62/49) * (.129 – ( (116/106) * .086 * (13/62)))
(r1 = .086/x1)
[1] 0.645
(r2 = ((116/106) * .086)/x2)
[1] 0.681
Hope that was clear. I just computed the pre-hiatus rates starting from 1950, since they didn’t give those in their table. 16/106 is an approximation of the effect of starting in 2000, vs 1998, and ending in 2012. They didn’t give that interval in their table S1.
If hiatus communication can be a proxy war, it seems naive not to expect bombastic announcements about the end or nonexistence of the hiatus just in time for the paper to be useful at the COP in Paris, rather than next year.
It’s also incomprehensible why a single paper should be considered as the last word and Truth about a scientific topic. The hiatus doesn’t end with Karl just as it didn’t continue with Roberts.
[…] Doug’s McNeall’s blog. […]
I’d argue that oversimplified binary statistical thinking is still better than an absence of (or hiatus in ;o) statistical thinking. If someone wants to make a claim based on observational data, then the scientific tradition is not to promulgate the claim unless the observations provide statistically significant evidence in support of the claim. In the case of the existence of a hiatus, there is evidence, but it is not statistically significant, so it is a mistake to claim unequivocally that there has been a hiatus *based solely on the data*.
“My colleague Chris Roberts points out that the necessary comparison is with the expected warming (and associated distribution) from the models, not from the somewhat arbitrary threshold of “zero” warming”
I’m not sure I really agree with the first part (although I definitely agree with the second part). Whether there has actually been a hiatus doesn’t depend on the models, but on the observations. The correct definition for me would be statistically significant evidence for a change in the rate of warming.
” There seems to have been a counter effort to demonstrate that the hiatus never existed in the first place. ”
I haven’t really noticed this, what I have noticed is people making an effort to demonstrate that there is no strong evidence that the apparent hiatus is anything other than the effects of natural variation, i.e. that the claim that there has been a slowdown in the underlying rate of warming is incorrect. That is not the same as saying the hiatus never existed.
AFAICS there is no statistically significant evidence for the existence of a hiatus, but there is no statistically significant evidence to rule out the existence of a hiatus (in the sense of a change in the underlying rate of warming), so we should keep an open mind and try to understand the physics.
As a statistician, I am much more influenced by physics than statistics.
“Whether there has actually been a hiatus doesn’t depend on the models, but on the observations. The correct definition for me would be statistically significant evidence for a change in the rate of warming.”
Whether or not the hiatus is defined in a model-independent way is a choice.
But doesn’t any calculation of statistical significance depend on a model, namely the model that you choose to describe your data under your null hypothesis?
The point I was making was that the putative hiatus may have an existence that is independent of our understanding of the physics. Sure the null hypothesis can be viewed as a model, but it is generally a statistical model, rather than a physical one (i.e. it only needs to explain what we expect to see under H0, but not why). It is perfectly reasonable to adopt a statistical definition of the hiatus (i.e. has there been a change in the linear trend – essentially a change-point analysis).
Introducing physical models introduces complications. If the trends in the models are inconsistent with the trends in the observations, that may mean that the models underestimate the plausible effects of internal variability (i.e. the unforced climate response), rather than because the models over-estimated the underlying rate of warming (i.e. the forced response of the climate). Or more probably a bit of both.
I think part of the problem is that people are not being clear about what they think the hiatus actually means. Is it a question of a reduction in the rate of surface warming (if so a change-point analysis is a good approach) or is it a question of a change in the forced response of the climate (in which case you would need a physical model as it is then a physics question rather than a statistical one). These are different problems and it is best to be clear on the question you want to ask when considering the definition of a hiatus, before choosing the form of the analysis.
[…] periods often discussed in terms of a ‘slowdown’ or ‘hiatus’. Carbon Brief, Doug McNeall & RealClimate all have excellent analyses of what this […]
“Climate skeptics have long overemphasised the hiatus – they’re unlikely to miss a good opportunity to claim that climate change isn’t happening, or that the models are wrong”
a. It is not about whether climate changes – of course it does. The key is to determine any manmade component from the background natural noise. Hadley pretended that had been done. Clearly they were wrong.
b. Since the models are all that were used to identify manmade warming from natural noise in the first place it is rather important that they get things right. Without the models there is no manmade signal detectable in the data and no way to predict the future.
c. The pause was expected – by sceptics – who had postulated an influence from the positive pdo. That being the case a negative pdo would lead to a slowdown, Trenberth (the pdo king) derided this notion initially but now it’s his main explanation for the pause – or was until he changed tack again with this new paper.
d. If the deep ocean is really hiding the warming without it being detected in the top 700m then it is just another natural variation. Clearly climate pessimists are now happy with such hand-waving but if sceptics had suggested such a mechanism for swallowing heat they’d 10 years ago they’d be called unscientific cranks.
e. Is Karl a pause denier? Is he anti-science for not believing the obs and just making his own up? Is he anti-consensus for disagreeing with the IPCC?
Isn’t this issue a scientific sideshow. Seems to me that the real issue is why the lower troposphere and radiosond data disagrees so much with the surface data and why the models seem to be predicting too much warming especially for the lower troposphere. Seems like a very important issue for resolving the convection/water vapor feedback issue.
Its not a sideshow. The “hiatus” is the same problem as the radiosond data. Neither fit the models. If the data and models do not agree then it is the models that are wrong, but I have never met a computer programmer yet who will accept that their code is wrong!
[…] do not expect changes in temperature to be uniform in space or time. We need to move away from the binary comparison of comparing individual short periods of temperature change and improve our understanding and […]