Congenital Cyclomania Redux

Guest Post by Willis Eschenbach

Well, I wasn’t going to mention this paper, but it seems to be getting some play in the blogosphere. Our friend Nicola Scafetta is back again, this time with a paper called “Solar and planetary oscillation control on climate change: hind-cast, forecast and a comparison with the CMIP5 GCMs”. He’s posted it up over at Tallbloke’s Talkshop. Since I’m banned over at Tallbloke’s, I thought I’d discuss it here. The paper itself is here, take your Dramamine before jumping on board. Dr. Scafetta has posted here on WUWT several times before, each time with his latest, greatest, new improved model. Here’s how well Scafetta’s even more latester, greatester new model hindcasts, as well as what it predicts, compared with HadCRUT4:

Figure 1. Figure 16A from Scafetta 2013. This shows his harmonic model alone (black), plus his model added to the average of the CMIP5 models following three different future “Representative Concentration Pathways”, or RCPs. The RCPs give various specified future concentrations of greenhouse gases. HadCRUT4 global surface temperature (GST) is in gray.

So far, in each of his previous three posts on WUWT, Dr. Scafetta has said that the Earth’s surface temperature is ruled by a different combination of cycles depending on the post:

First Post: 20 and 60 year cycles. These were supposed to be related to some astronomical cycles which were never made clear, albeit there was much mumbling about Jupiter and Saturn.

Second Post: 9.1, 10-11, 20 and 60 year cycles. Here are the claims made for these cycles:

9.1 years : this was justified as being sort of near to a calculation of (2X+Y)/4, where X and Y are lunar precession cycles,

“10-11″ years: he never said where he got this one, or why it’s so vague.

20 years: supposedly close to an average of the sun’s barycentric velocity period.

60 years: kinda like three times the synodic period of Jupiter/Saturn. Why three times? Why not?

Third Post:  9.98, 10.9, and 11.86 year cycles. These are claimed to be

9.98 years: slightly different from a long-term average of the spring tidal period of Jupiter and Saturn.

10.9 years: may be related to a quasi 11-year solar cycle … or not.

11.86 years: Jupiter’s sidereal period.

The latest post, however, is simply unbeatable. It has no less than six different cycles, with periods of 9.1, 10.2, 21, 61, 115, and 983 years. I haven’t dared inquire too closely as to the antecedents of those choices, although I do love the “3” in the 983 year cycle. Plus there’s a mystery ingredient, of course.

Seriously, he’s adding together six different cycles. Órale, that’s a lot! Now, each of those cycles has three different parameters that totally define the cycle. These are the period (wavelength), the amplitude (size), and the phase (starting point in time) of the cycle.

This means that not only is Scafetta exercising free choice in the number of cycles that he includes (in this case six). He also has free choice over the three parameters for each cycle (period, amplitude, and phase). That gives him no less than 18 separate tunable parameters.

Just roll that around in your mouth and taste it, “eighteen tunable parameters”. Is there anything that you couldn’t hindcast given 18 different tunable parameters?

Anyhow, if I were handing out awards, I’d certainly give him the first award for having eighteen arbitrary parameters. But then, I’d have to give him another award for his mystery ingredient.

Because of all things, the mystery ingredient in Scafetta’s equation is the average hindcast (and forecast) modeled temperature of the CMIP5 climate models. Plus the mystery ingredient comes with its own amplitude parameter (0.45), along with a hidden parameter for the zero point of the average model temperatures before being multiplied by the amplitude parameter. So that makes twenty different adjustable parameters.

Now, I don’t even know what to say about this method. I’m dumbfounded. He’s starting with the average of the CMIP5 climate models, adjusted by an amplitude parameter and a zeroing parameter. Then he’s figuring the deviations from that adjusted average model result based on his separate 6-cycle, 18-parameter model. The sum of the two is his prediction. I truly lack words to describe that, it’s such an awesome logical jump I can only shake my head in awe at the daring trapeze leaps of faith …

I suppose at this point I need to quote the story again of Freeman Dyson, Enrico Fermi, “Johnny” Von Neumann, and the elephant. Here is Freeman Dyson, with the tale of tragedy:

By the spring of 1953, after heroic efforts, we had plotted theoretical graphs of meson–proton scattering.We joyfully observed that our calculated numbers agreed pretty well with Fermi’s measured numbers. So I made an appointment to meet with Fermi and show him our results. Proudly, I rode the Greyhound bus from Ithaca to Chicago with a package of our theoretical graphs to show to Fermi.

When I arrived in Fermi’s office, I handed the graphs to Fermi, but he hardly glanced at them. He invited me to sit down, and  asked me in a friendly way about the health of my wife and our newborn baby son, now fifty years old. Then he delivered his verdict in a quiet, even voice.

“There are two ways of doing calculations in theoretical physics”, he said. “One way, and this is the way I prefer, is to have a clear physical picture of the process that you are calculating. The other way is to have a precise and self-consistent mathematical formalism. You have neither.”

I was slightly stunned, but ventured to ask him why he did not consider the pseudoscalar meson theory to be a self-consistent mathematical formalism. He replied, “Quantum electrodynamics is a good theory because the forces are weak, and when the formalism is ambiguous we have a clear physical picture to guide us.With the pseudoscalar meson theory there is no physical picture, and the forces are so strong that nothing converges. To reach your calculated results, you had to introduce arbitrary cut-off procedures that are not based either on solid physics or on solid mathematics.”

In desperation I asked Fermi whether he was not impressed by the agreement between our calculated numbers and his measured numbers. He replied, “How many arbitrary parameters did you use for your calculations?” I thought for a moment about our cut-off procedures and said, “Four.” He said, “I remember my friend Johnny von Neumann used to say, with four parameters I can fit an elephant, and with five I can make him wiggle his trunk.”

With that, the conversation was over. I thanked Fermi for his time and trouble, and sadly took the next bus back to Ithaca to tell the bad news to the students.

Given that lesson from Dyson, and bearing in mind that Scafetta is using a total of 20 arbitrary parameters … are we supposed to be surprised that Nicola can make an elephant wiggle his trunk? Heck, with that many parameters, he should be able to make that sucker tap dance and spit pickle juice …

Now, you can expect that if Nicola Scafetta shows up, he will argue that somehow the 20 different parameters are not arbitrary, oh, no, they are fixed by the celestial processes. They will likely put forward the same kind of half-ast-ronomical explanation  they’ve used before—that this one represents (2X+Y)/4, where X and Y are lunar precession cycles, or that another one’s 60 year cycle is kind of near three times the synodic period of Jupiter and Saturn (59.5766 years) and close is good enough, that kind of thing. Or perhaps they’ll make the argument that Fourier analysis shows peaks that are sort of near to their chosen numbers, and that’s all that’s needed.

The reality is, if you give me a period in years, I can soon come up with several astronomical cycles that can be added, subtracted, and divided to give you something very near the period you’ve given me … which proves nothing.

Scafetta has free choice of how many cycles to include, and free choice as to the length, amplitude, and phase of each those cycles. And even if he can show that the length of one of his cycles is EXACTLY equal to some astronomical constant, not just kind of near it, he still has totally free choice of phase and amplitude for that cycle. So to date, he’s the leading contender for the 2013 Johnny Von Neumann award, which is given for the most tunable parameters in any scientific study.

The other award I’d give this paper would be for Scafetta’s magical Figure 11, which I reproduce below in all its original glory.

Figure 2. Scafetta’s Figure 11 (click to enlarge) ORIGINAL CAPTION: (Left) Schematic representation of the rise and fall of several civilizations since Neolithic times that well correlates with the 14C radio- nucleotide records used for estimating solar activity (adapted from Eddy’s figures in Refs. [90, 91]). Correlated solar-climate multisecular and millennial patterns are recently confirmed [43, 44, 47]. (Right) Kepler’s Trigon diagram of the great Jupiter and Saturn conjunctions between 1583 to 1763 [89], highlighting 20 year and 60 year astronomical cycles, and a slow millennial rotation. 

First off, does that graphic, Figure 11 in Scafetta’s opus, make you feel better or worse about Dr. Scafetta’s claims? Does it give you that warm fuzzy feeling about his science? And why are Kepler’s features smooched out sideways and his fingers so long? At least let me give the poor fellow back his original physiognomy.

There, that’s better. Next, you need to consider the stepwise changes he shows in “carbon 14”, and the square-wave nature of the advance and retreat of alpine glaciers at the lower left. That in itself was good, I hadn’t realized that the glaciers advanced and retreated in that regular a fashion, or that carbon 14 was unchanged for years before and after each shift in concentration. And I did appreciate that there were no units for any of the four separate graphs on the page, that counted heavily in his favor. But what I awarded him full style points for was the seamless segue from alpine glaciers to the “winter severity index” in the year 1000 … that was a breathtaking leap.

And as you might expect from a man citing Kepler, Scafetta treats scientific information like fine wine—he doesn’t want anything of recent vintage. Apparently on his planet you have to let science mellow for some decades before you bring it out to breathe … and in that regard, I direct your attention to the citation in the bottom center of his Figure 11, “Source: Geophysical Data, J. Biddy J. B. Eddy (USA) 1978″. (Thanks to Nicola for the correction, the print was too small to read.)

Where he stepped up to the big leagues, though, is in the top line in the chart. Click on the chart to enlarge it if you haven’t done so yet, so you can see all the amazing details. The “Sumeric Maximum”, the collapse of Machu Pichu, the “Greek Minimum”, the end of the Maya civilization, the “Pyramid Maximum” … talk about being “Homeric in scope”, he’s even got the “Homeric Minimum”.

Finally, he highlights the “20 year and 60 year astronomical cycles” in Kepler’s chart at the right. In fact, what he calls the “20 year” cycles shown in Kepler’s dates at the right vary from 10 to 30 years according to Kepler’s own figures shown inside the circle, and what he calls the “60 year astronomical cycles” include cycles from 50 to 70 years …

In any case, I’m posting all of this because I just thought folks might like to know of Nicola Scafetta’s latest stunning success. Using a mere six cycles and only twenty tunable parameters plus the average of a bunch of climate models, he has emulated the historical record with pretty darn good accuracy.

…

And now that he has explained just exactly how to predict the climate into the future, I guess the only mystery left is what he’ll do for an encore performance. Because this most recent paper of his, this one will be very hard to top.

In all seriousness, however, let me make my position clear.

Are there cycles in the climate? Yes, there are cycles. However, they are not regular, clockwork cycles like those of Jupiter and Saturn. Instead, one cycle will appear, and will be around for a while, and then disappear to be replaced by some longer or shorter cycle. It is maddening, frustrating, but that’s the chaotic nature of the beast. The Pacific Decadal Oscillation doesn’t beat like a clock, nor does the El Nino or the Madden-Julian oscillation or any other climate phenomena.

What is the longest cycle that can be detected in a hundred year dataset? My rule of thumb is that even if I have two full cycles, my results are too uncertain to lean on. I want three cycles so I can at least get a sense about the variation. So for a hundred year dataset, any cycle over fifty years in length is a non-starter, and thirty-three years and shorter is what I will start to trust.

Can you successfully hindcast temperatures using other cycles than the ones Scafetta uses? Certainly. He has demonstrated that himself, as this is the fourth combination of arbitrarily chosen cycles that he has used. Note that in each case he has claimed the model was successful. This by no means exhausts the possible cycle combinations that can successfully emulate the historical temperature.

Does Scafetta’s accomplishment mean anything? Sure. It means that with six cycles and no less than twenty tunable parameters, you can do just about anything. Other than that, no. It is meaningless.

Could he actually test his findings? Sure, and I’ve suggested it to him. What you need to do is run the analysis again, but this time using the data from say 1910 to 1959 only. Derive your 20 fitted variables using this data alone.

Then test your 20 fitted variables against the data from 1960 to 2009, and see how the variables pan out.

Then do it the other way around. Train the model on the later data, and see how well it does on the early data. It’s not hard to do. He knows how to do it. But if he has ever done it, I have not seen anywhere that he has reported the results.

How do I know all this? Folks, I can’t tell you how many late nights I’ve spent trying to fit any number and combination of cycles to the historical climate data. I’ve used Fourier analysis and periodicity analysis and machine-learning algorithms and wavelets and stuff I’ve invented myself. Whenever I’ve thought I have something, as soon as it leaves the training data and starts on the out-of-sample data, it starts to diverge from reality. And of course, the divergence increases over time.

But that’s simply the same truth we all know about computer weather forecasting programs—out-of-sample, they don’t do all that well, and quickly become little better than a coin flip.

Finally, even if the cycles fit the data and we ignore the ridiculous number of arbitrary parameters, where is the physical mechanism connecting some (2*X+T)/4 combination of two astronomical cycles, and the climate? As Enrico Fermi pointed out, you need to have either “a clear physical picture of the process that you are calculating” or “a precise and self-consistent mathematical formalism”. 

w.

PS—Please don’t write in to say that although Nicola is wrong, you have the proper combination of cycles, based on your special calculations. Also, please don’t try to explain how a cycle of 21 years is really, really similar to the Jupiter-Saturn synodic cycle of 19+ years. I’m not buying cycles of any kind, motorcycles, epicycles, solar cycles, bicycles, circadian cycles, nothing. Sorry. Save them for some other post, they won’t go bad, but please don’t post them here.