A great deal of discussion lately has centered around the idea of a recent “slowdown” in global temperature. With 2016 destined to break the hottest-year-ever record for the third time in a row, it would seem to be over now, even if it was real. I say “if” because scientists disagree about whether or not it ever was even a real thing — and I’m one of those who thinks it was not.
A new paper in Nature Communications investigates the likelihood of a “prolonged slowdown in global warming in the early 21st century.” But like most (if not all) papers that discuss the so-called “slowdown” it does nothing to establish that such a slowdown is real, that it was anything but random fluctuation that looks like a slowdown. Those who do statistics, and do it right, learned one of its most important lessons a long, long time ago: that “looks like” is a very bad way to draw conclusions.
Their central theme is that from 1998 through 2015 global surface temperature rose at a rate of merely “about” 0.1 °C per decade, and that using estimated multi-decadal variability from their simulation model (at the Geophysical Fluid Dynamics Laboratory) there’s a non-trivial possibility this could persist, even until 2030. They use only one computer model, one with particularly high multidecadal variability, but I’m not qualified to comment on that. But the paper starts from more than one mistaken premise. First, that the rate of “global warming” was a mere 0.1 °C per decade, and second, that we need anything new to understand the observed variations in global surface temperature or its estimated rates of increase.
Let’s start with that “0.1 °C per decade” thing. Setting aside the fact that referring (in the paper’s title) to “global warming” quite ignores all the other things besides just surface temperature that have changed during that time span, we can compute the estimated warming rate and its uncertainty (something the authors don’t bother to do) from 1998 through 2015 for the five best-known surface temperature data sets (from NASA, NOAA, the Hadley Centre/Climate Research Unit, Cowtan & Way, and the Berkeley Earth Surface Temperature project):
I find it rather troubling that they consistently refer to the warming rate as “about 0.1” when all five data sets give a central estimate higher than that. It’s true that they all include 0.1 in their (roughly 95%) confidence intervals — but four of the five also include 0.2 (their stated rate for climate models). The fact that 4 out of 5 include both limits is a sign — a sign that the confidence intervals are very wide — and should be taken as a sign that making flat statements about “slowdown,” let alone quoting a numerical figure lower than all five central estimates — is statistically “dodgy” at best.
They also mention that using the time span from 1998 through 2015 enables them to include el Niño peaks at both ends. But the real peak of the recent el Niño isn’t 2015, it’s 2016. And what happens if we estimate trend rates for the time span 1998 through 2016 (using the year-so-far values) rather than through 2015? We get this:
If we bracket the time span with both actual el Niño peaks, then 4 out of 5 data sets give a central estimate which is closer to their standard of 0.2 °C per decade than their standard of 0.1 °C per decade.
Which makes one wonder: what is the likelihood of the purported “slowdown” lasting 15 years longer (through 2030), when it can’t even last one year longer?
Then there’s the “broken trend” issue. Looking at data from 1975 through 2015, and using the data set with the lowest warming rate 1998-through-2015 (HadCRU), here’s the trend model required to claim a “slowdown” during the 1998-through-2015 period:
This model doesn’t just have a lower central estimate of slope, it has a sudden jump as well, one nearly as big as what they claim for the rate per decade during the latter time span. If you want to “explain” a lower rate of increase (without bothering to demonstrate that it really exists), perhaps you should begin by explaining why global surface temperature started that time span with a jump nearly as big as your claimed decadal rate of increase.
We can fit more realistic models, with a rate change during the period from 1998 onward but which do not include a jump discontinuity. Let’s not omit 2016, since the whole subject is the possibility of a sustained episode of “slowdown” (possibly all the way through 2030). What warming rates would we get for the 1998-onward period? These:
Confidence intervals for all five data sets include 0.2 °C per decade, their stated value of model averages, but all five exclude 0.1, their stated value for “prolongued slowdown.” How can one discuss the possibility of a prolongued slowdown, when the data we have already contradict it?
As for the variability in warming rates (whether statistically meaningful or not), do we really need to invoke multi-decadal variability to explain it? We can take each data set and remove an estimate of variability caused by much more short-term factors, namely el Niño, volcanic eruptions, and solar variations (using a method like that from Foster & Rahmstorf). The data corrected for short-term fluctuation factors look like this:
The dashed lines show their period of claimed “slowdown” — but it not only fails statistical tests, it doesn’t even pass the “looks like” test.
In my opinion, this paper doesn’t pass muster. They discuss a “slowdown” without providing any real evidence of it, they fail to estimate confidence limits for their “slowdown” rate which happens to be lower than the central estimate for all five major global data sets, they ignore the presence of a jump discontinuity in the model required to get such low estimates, they fail to mention that even the variability which does exist is readily explained by short-term factors without having to invoke multi-decadal variability, and their central thesis — that a “slowdown” might persist for another 15 years — is already contradicted by a single additional year. I consider its many, and very serious, flaws to be telling evidence that the whole “slowdown” idea was misguided from the very start.
This blog is made possible by readers like you; join others by donating at Peaseblossom’s Closet.