Guest Post by Willis Eschenbach
There’s a new paper published in Nature Scientific Reports called “Identification of the driving forces of climate change using the longest instrumental temperature record”, by Geli Wang et al, hereinafter Wang2017.
By “the longest instrumental temperature record” they mean the Central England Temperature, commonly called the “CET”. Unfortunately, the CET is a spliced, averaged, and adjusted temperature record. Not only that, but the underlying datasets from which it is constructed have changed over time. Here are some details from the study by Parker et. al.
Between 1772 and 1876 the daily series is based on a sequence of single stations whose variance has been reduced to counter the artificial increase that results from sampling single locations. For subsequent years, the series has been produced from combinations of as few stations as can reliably represent central England in the manner defined by Manley. We have used the daily series to update Manley’s published monthly series in a consistent way.
We have evaluated recent urban warming influences at the chosen stations by comparison with nearby rural stations, and have corrected the series from 1974 onwards.
According to the paper, no less than 14 different and distinct datasets were used to construct the CET.
Perhaps predictably, the authors of Wang2017 completely fail to mention any of this … instead, they simply say:
As the world’s longest instrumental record of temperature, the Met Office Hadley Centre’s CET time series represents the monthly mean surface air temperature averaged over the English midlands and spans the period January 1659 to December 2013.
Well … no, not really. And more to the point, using such a spliced, averaged, and adjusted dataset for an analysis of the underlying “driving forces” is totally inappropriate.
Now, in the Wang2017 analysis, they claim to find a couple of “driving forces” of the CET. Of these they say:
The peak L1 = 3.36 years seems to empirically correspond to the El Niño-Southern Oscillation (ENSO) signal, which has a period range of within 3 to 6 years. ENSO is arguably the most important global climate pattern and the dominant mode of climate variability13. The effect of ENSO on climate in Europe has been studied intensively using both models and observational or proxy data e.g. refs 14, 15, and a consistent and statistically significant ENSO signal on the European climate has been found e.g. refs 14 and 16.
The peak L4 = 22.6 years is coincident with the Hale sunspot cycle.
Let me start by saying that a claim that something “seems to empirically correspond” with something else is not a falsifiable claim … and that means it is not a scientific claim in any sense. And the same is true for a claim that something “is coincident with” something else. The use of such terms is scientific doublespeak, bafflegab of the highest order.
Setting that aside, here’s what the CET actually looks like:
Now, there is a commonly-used way to determine whether two datasets are related. This is a cross-correlation analysis, which shows more than just the correlation of the two datasets. It shows the correlation at various lag and lead times. Here is the cross-correlation analysis of the Central England Temperature and the El Nino datasets:
Does the El Nino affect the temperature in Central England? Well, perhaps, with a half-year lag or so. But it’s a very, very weak effect.
Then we have their claim about the relationship of the CET with sunspots, wherein they make the claim that a 22.6-year signal is “coincident with” the sun’s Hale Cycle. “Coincident with” … sheesh …
Now, the “Hale Cycle” reflects the fact that around the time of the sunspot maximum, the magnetic polarity of the sun reverses. As a result, the Hale Cycle is the length of time from any given sunspot peak to the peak of the second sunspot cycle following the given peak.
And how long is the Hale Cycle? Well, here’s a histogram of the different lengths, from NASA data …
So … is a signal with a 22.6-year cycle “coincident with” the length of the Hale Cycle? Well, sure … but the same is true of any cycle length from 17 to 28 years. Color me totally unimpressed.
Next, do the sunspots actually affect the temperature of Central England? Again, the cross-correlation function comes into play:
Basic answer is … well … no. Cross-correlation shows no evidence that the sunspots have any significant effect on the CET.
Finally, what kinds of signals do show up in the CET data? To answer this question, I use the Complete Ensemble Empirical Mode Decomposition method, as discussed here. Below, I show the CEEMD decomposition of the monthly CET data. The upper graph (blue) shows the different empirical mode cycles (C1 to C7) which when added together along with the Residual give us the raw data in the top panel.
The lower graph (red) shows the periodogram of each of those same empirical mode cycles.
Of all of these empirical modes, the strongest signal is at about 15 years (C4, lower red graph). There is a signal at about 24 years (C5, lower red graph), but it is much weaker, less than half the strength. In the corresponding mode C5 in the upper blue graph we can see why—sometimes we can see a cycle in the 25 to 30-year range, but it fades in and out.
To me, this is one big advantage of the CEEMD method—it shows not only the strength of the various cycle lengths, but also just where in the entire timespan of the dataset the cycles are strong, weak, or non-existent. This lets us see whether we are looking at a real persistent cycle which is visible from the start to the finish of the data, or whether it is a pseudo-cycle which fades in and out of existence.
Finally, is there any evidence of anthropogenic global warming in the CET data? To answer this, look at the residual signal in the bottom panel of the blue graph above. This is what remains after all of the underlying cyclical waves have been removed … looking at that it seems that there is no unusual recent warming of any kind.
My conclusion? If you use enough different statistical methods, as Wang2017 has done, you can dig out even the most trivial underlying cycles of a dataset … but the reality is, when you decompose even the most random of signals, it will show peaks in the underlying cycles. It has to—as Joe Fourier showed, every signal can be decomposed into underlying simpler waves. However, this does not mean that those underlying simpler waves have any kind of meaning or significance.
Finally, an oddity. Look at Mode C2 in the upper blue graph. I suspect that those blips are related to the spliced nature of the CET dataset. When you splice two datasets together, it seems to me that you’ll get some kind of higher-frequency “ringing” around the splice. Below I show Mode C2 along with what I can determine regarding the dates of the splices in the CET …
Is this probative of the theory that these are related to the splices? By no means … but it is certainly suggestive.
Here on the coast north of San Francisco, after two days of one hundred plus temperatures the clouds and fog are returning, and the evening is cool … what a world.
My best to everyone, in warm and cool climes alike,
My Usual Request: When you comment, please QUOTE THE EXACT WORDS THAT YOU ARE DISCUSSING, so that we can all understand just what you are referring to.
My Other Request: Please do not stop after merely claiming I’m using the wrong dataset or the wrong method. I may well be wrong, but such observations are not meaningful unless and until you add a link to the proper dataset or give us a demonstration of the right method.