Guest Post by Willis Eschenbach
I thought I was done with sunspots … but as the well-known climate scientist Michael Corleone once remarked, “Just when I thought I was out … they pull me back in”. In this case Marcel Crok, the well-known Dutch climate writer, asked me if I’d seen the paper from Nir Shaviv called “Using the Oceans as a Calorimeter to Quantify the Solar Radiative Forcing”, available here. Dr. Shaviv’s paper claims that both the ocean heat content and the ocean sea surface temperature (SST) vary in step with the ~11 year solar cycle. Although it’s not clear what “we” means when he uses it, he says: 
“We find that the total radiative forcing associated with solar cycles variations is about 5 to 7 times larger than just those associated with the TSI variations, thus implying the necessary existence of an amplification mechanism, though without pointing to which one.” Since the ocean heat content data is both spotty and incomplete, I looked to see if the much more extensive SST data actually showed signs of the claimed solar-related variation.
To start with, here’s what Shaviv2008 says about the treatment of the data:
Before deriving the global heat flux from the observed ocean heat content, it is worth while to study in more detail the different data sets we used, and in particular, to better understand their limitations. Since we wish to compare them to each other, we begin by creating comparable data sets, with the same resolution and time range. Thus, we down sample higher resolution data into one year bins and truncate all data sets to the range of 1955 to 2003.
I assume the 1955 start of their data is because the ocean heat content data starts in 1955. Their study uses the HadISST dataset, the “Ice and Sea Surface Temperature” data, so I went to the marvelous KNMI site and got that data to compare to the sunspot data. Here are the untruncated versions of the SIDC sunspot and the HadISST sea surface temperature data.
Figure 1. Sunspot numbers (upper panel) and sea surface temperatures (lower panel).
So … is there a solar component to the SST data? Well, looking at Figure 1, for starters we can say that if there is a solar component to SST, it’s pretty small. How small? Well, for that we need the math. I often start with a cross-correlation. A cross-correlation looks not only at how well correlated two datasets might be. It also shows how well correlated the two datasets are with a lag between the two. Figure 2 shows the cross-correlation between the sunspots and the SST:
Figure 2. Cross-correlation, sunspots and sea surface temperatures. Note that they are not significant at any lag, and that’s without accounting for autocorrelation.
So … I’m not seeing anything significant in the cross-correlation over full overlap of the two datasets, which is the period 1870-2013. However, they haven’t used the full dataset, only the part from 1955 to 2003. That’s only 49 years … and right then I start getting nervous. Remember, we’re looking for an 11-year cycle. So results from that particular half-century of data only represent three complete solar cycles, and that’s skinny … but in any case, here’s cross-correlation on the truncated datasets 1955-2003:
Figure 3. Cross-correlation, truncated sunspots and sea surface temperatures 1955-2003. Note that while they are larger than for the full dataset, they are still not significant at any lag, and that’s without accounting for autocorrelation.
Well, I can see how if all you looked at was the shortened datasets you might believe that there is a correlation between SST and sunspots. Figure 3 at least shows a positive correlation with no lag, one which is almost statistically significant if you ignore autocorrelation.
But remember, in the cross-correlation of the complete dataset shown back in Figure 2, the no-lag correlation is … well … zero. The apparent correlation shown in the half-century dataset disappears entirely when we look at the full 140-year dataset.
This highlights a huge recurring problem with analyzing natural datasets and looking for regular cycles. Regular cycles which are apparently real appear, last for a half century or even a century, and then disappear for a century …
Now, in Shaviv2008, the author suggests a way around this conundrum, viz:
Another way of visualizing the results, is to fold the data over the 11-year solar cycle and average. This reduces the relative contribution of sources uncorrelated with the solar activity as they will tend to average out (whether they are real or noise).
In support of this claim, he shows the following figure:
Figure 4. This shows Figure 5 from the Shaviv2008 paper. Of interest to this post is the top panel, showing the ostensible variation in the averaged cycles.
Now, I’ve used this technique myself. However, if I were to do it, I wouldn’t do it the way he has. He has aligned the solar minimum at time t=0, and then averaged the data for the 11 years after that. If I were doing it, I think I’d align them at the peak, and then take the averages for say six years on either side of the peak.
But in any case, rather than do it my way, I figured I’d see if I could emulate his results. Unfortunately, I ran into some issues right away when I started to do the actual calculations. Here’s the first issue:
Figure 5. The data used in Shaviv2008 to show the putative sunspot-SST relationship.
I’m sure you can see the problem. Because the dataset is so short (n = 49 years), there are only four solar minima—1964, 1976, 1986, and 1996. And since the truncated data ends in 2003, that means that we only have three complete solar cycles during the period.
This leads directly to a second problem, which is the size of the uncertainty of the results of the “folded” data. With only three full cycles to analyze, the uncertainty gets quite large. Here are the three folded datasets, along with the mean and the 95% confidence interval on the mean.
Figure 6. Sea surface temperatures from three full solar cycles, “folded” over the 11-year solar cycle as described in Shaviv2008
Now, when I’m looking for a repetitive cycle, I look at the 95% confidence interval of the mean. If the 95%CI includes the zero line, it means the variation is not significant. The problem in Figure 6, of course, is the fact that there are only three cycles in the dataset. As a result, the 95%CI goes “from the floor to the ceiling”, as the saying goes, and the results are not significant in the slightest.
So why does the Shaviv2008 result shown in Figure 4 look so convincing? Well … it’s because he’s only showing one standard error as the uncertainty in his results, when what is relevant is the 95%CI. If he showed the 95%CI, it would be obvious that the results are not significant.
However, none of that matters. Why not? Well, because the claimed effect disappears when we use the full SST and sunspot datasets. Their common period goes from 1870 through 2013, so there are many more cycles to average. Figure 7 shows the same type of “folded” analysis, except this time for the full period 1870-2013:
Figure 7. Sea surface temperatures from all solar cycles from 1870-2013, “folded” over the 11-year solar cycle as described in Shaviv2008
Here, we see the same thing that was revealed by the cross-correlation. The apparent cycle that seemed to be present in the most recent half-century of the data, the apparent cycle that is shown in Shaviv2008, that cycle disappears entirely when we look at the full dataset. And despite having a much narrower 95%CI because we have more data, once again there is no statistically significant departure from zero. At no time do we see anything unexplainable or unusual at all
And so once again, I find that the claims of a connection between the sun and climate evaporate when they are examined closely.
Let me be clear about what I am saying and not saying here. I am NOT saying that the sun doesn’t affect the climate.
What I am saying is that I still haven’t found any convincing sign of the ~11-year sunspot cycle in any climate dataset, nor has anyone pointed out such a dataset. And without that, it’s very hard to believe that even smaller secular variations in solar strength can have a significant effect on the climate.
So, for what I hope will be the final time, let me put out the challenge once again. Where is the climate dataset that shows the ~11-year sunspot/magnetism/cosmic rays/solar wind cycle? Shaviv echoes many others when he claims that there is some unknown amplification mechanism that makes the effects “about 5 to 7 times larger than just those associated with the TSI variations” … however, I’m not seeing it. So where can we find this mystery ~11-year cycle?
Please use whatever kind of analysis you prefer to demonstrate the putative 11-year cycle—”folded” analysis as above, cross-correlation, wavelet analysis, whatever.
Regards,
w.
My Usual Request: If you disagree with someone, myself included, please QUOTE THE EXACT WORDS YOU DISAGREE WITH. This prevents many flavors of misunderstanding, and lets us all see just what it is that you think is incorrect.
Subject: This post is about the quest for the 11-year solar cycle. It is not about your pet theory about 19.8 year Jupiter/Saturn synoptic cycles. If you wish to write about them, this is not the place. Take it to Tallbloke’s Talkshop, they enjoy discussing those kinds of cycles. Here, I’m looking for the 11-year sunspot cycles in weather data, so let me ask you kindly to restrict your comments to subjects involving those cycles.
Data and Code: I’ve put the sunspot and HadISST annual data online, along with the R computer code, in a single zipped folder called “Shaviv Folder.zip“
Whoops. Re previous comment, “second harmonic” was in error, that’s not frequency*2, not an overtone. Although the link is still informative reading. It’d actually be the 1/2 subharmonic, frequency divided by 2.
But you get the idea. What happens every 24 years may look like it happens every 48 years, or 72 years, etc.
the problems with this thread-
mosher thinking-
“The solar variation is so minor that I do the following thought experiment.
I make a chart of TSI versus time.
I label it c02.
I make a chart of temperature versus time
I label it temperature.
Then I imagine what a skeptic would say If I claimed that chart one helped to explain chart two.”
make it about something else, then ignore the obvious things like … ohh, just about every proxy we have that says co2 lags temps, NOT the other way around AND temps move with sunspot numbers through Be records. hmm, it takes skill to ignore so much.
this thread is a fine example of that skill in action. willis congrats mosher for thinking like him, which says it all really.
it is obvious that the goal of this thread is not to determine whether there is an eleven year cycle evident in the record, it is just to belittle those looking for it. it really is not a simple task. there are just so many possibilies there. to claim that one knows without doubt that it should be most visible in the the 11 year cycle is a clear failure and outright arrogance. it is this form of arrogance that we see every day in the climate science community. it is unexpected from the engineering community, which i thought willis was from.
did willis die recently and mosher take over his account?!
mobihci: link please to your Be correlation. Better yet, download the data sets and see for yourself.
richard verney says: Konrad says…
“We hold similar views, and recall that we engaged in much discussion on this on Willis’ article on ‘Radiating the Oceans’ (which i personally consider to be not one of Willis’ stronger articles – sorry willis, just my personal opinion). However, the point that the warmists would raise is that if DWLWIR heats the atmosphere, even if it does not heat the ocean below, then due to the warmer atmosphere above the ocean,the heat loss from the ocean is lower/slower, thereby helping to maintain or even produce higher ocean temperatures over time.”
Just complimenting you guys along with Greg on some good insights and cogent explanations. I did want to point out that the mention of spray and spume and the generally wet conditions above the ocean are only part of the LWIR absorption. Water vapour itself didn’t get a direct mention.
Something else worth mentioning is that ocean water is quite reflective as well as having a high emissivity. I do not believe the story about water having an E of 0.7 – I have measured too much water with an IR thermometer to accept such a low number. The emissivity of water is in the high ‘9’s’. A common mistake would be to measure the water’s bulk temperature and interpret that as being the same as the surface skin temperature. It is about the same as black oil which is also quite reflective though they react differently to the angle of incidence.
The big influence is of course the water vapour near the ocean. I don’t think much LW gets to the surface. What leaves bounces back and forth a lot too of course, so overall the vapour is an insulating blanket. CO2 is simply a bit player there is so little of it.
A friend sent me a message the other day saying that the volume of water condensed from vapour created per annum by burning fossil fuels is about the same as the volume of Lake Simcoe in Ontario. That’s a lot of GHG but also a lot of condensing heat transport medium. The thunderstorm thermostat hypothesis rules, OK?
“Where is the climate dataset that shows the ~11-year sunspot/magnetism/cosmic rays/solar wind cycle?”
The solar wind is not in a well defined ~11 cycle: http://snag.gy/hSqT4.jpg
Greg Goodman says:
June 8, 2014 at 7:47 am
“IMO, it is likely that the various long period climate “oscillations” are the result of interactions of lunar and solar influences with periods around 9y and 10-11y respectively.”
I agree. I do not see a peak in the SST PSD specifically associated with 11 years, but there are several peaks at harmonics corresponding to such interactions, and in measure of what might be expected if they modulate one another.
I don’t want to get mired again in non-productive dialog with Willis. But, I did happen upon this entry in Wikipedia which may be of interest to him. Perhaps it has already been pointed out to him, and I’ll probably get blasted for some sort of imagined deviousness in providing it. For the record, I’m not trying to make any point at all, just trying to be helpful, as always.
“As a result, there is no “un-even solar heating of the polar regions due to the distance from the Sun”.It doesn’t exist.”
Annually no, but seasonally there is.
Greg Goodman says:
June 7, 2014 at 10:48 pm
“Forgot to answer”? I never even noticed the question. As usual, your assumed omniscience doesn’t do you credit. You don’t know what other people know or have forgotten, and your assumption that you do know is neither pleasant nor conducive to discussion.
Naw, let’s worry about it. You proposed the bizarre procedure of doing a fourier transform of a cross-correlation function, and made ludicrous claims about it. I showed that it reveals nothing about the target dataset, and that you’d get similar results from red noise. Now you suddenly want to forget about it … as would I in your place. But we digress …
Glad to. That is not “my” 0.2 threshold. It is the calculated level for statistical significance at p = 0.05, as calculated by the program, without adjustment for autocorrelation. What do you use for statistical significance of a cross-correlation, if not that?
No clue. You haven’t indicated anywhere what exact data was used to make the graph, nor how you made it. Since the significance level is a function of both the number of data points and the standard deviation of the two datasets, there’s not enough data to answer your question.
What your cited graph actually looks like, however, is that you forgot to detrend your datasets before doing the cross-correlation …
w.
Peter Sable says:
June 7, 2014 at 10:43 pm
Peter, whenever like you someone starts bitching and whining about people’s choice of computer language, and starts claiming great statistical insight into the errors of others without identifying those errors, it’s a clear sign that they can’t find a problem in my computer code or logic, so they are grasping at straws … Peter, if you have a problem with my work or the work of others, either specify it or go away. This kind of boastful attack you’ve put on here is as far from science as you can get.
w.
michaelwiseguy says:
June 8, 2014 at 12:10 am
Thanks, Michael. We can find the daily solar signal and the annual signal with no problem at all, despite the 352,670,000,000,000,000,000 gallons of water. And people keep claiming that they can find 60-year and 100-year solar cycles in the ocean temperatures (although I’ve not seen them).
Finally, and most relevant to this post, Nir Shaviv has claimed very strongly that he can find the 11-year signal in the SST (although I can’t find a significant signal)
So it’s not clear why you think an 11-year signal would be hard to find.
w.
Alex says:
June 8, 2014 at 1:48 am
I love these charming folks that like to make their departure note as nasty and vicious as possible. Somehow, they are under the illusion that this makes them look … what? Look smarter? Look better? Look strong and decisive?
Mostly it just makes them look terribly needy. Why else would they care what we thought about their leaving? My goodness, what if we didn’t notice that they had left, how terrible that would be! Better to go out with a bang …
Unsurprisingly, it’s usually the anonymous interchangeable internet popups like Alex who do it, the folks that are unwilling to sign their own name and own their own words.
Fortunately, it’s a lovely morning outside, and last night the gorgeous ex-fiancee and I got to watch a couple of fox kits gamboling and playing on our porch … the world is good.
w.
Willis, if you reply to on phrase in a comment, I am assuming telepathic powers by assuming you read to following line as well.
Greg Goodman says:
June 7, 2014 at 12:56 pm
“So smart I’d done it and posted it 9 hours before you told me “directly” to get off my butt and do it.
I’ll take that as an almost apology 😉
Without worrying about the FT of CC for the moment. What information do you think can be derived from cross-correlation. You used it to support your impression that there is no solar signal, so you must expect that something could be there that was not. ”
===
But I digress. Let’s continue the search for the illusive solar signal.
I used the full hadISST monthly data from the link shown in the graph. The SSN data was cropped to the same starting date (1870). The cross-correlation is the cross-correlation of the full overlap available at each lag ( as I’ve pointed out before a flat line is not valid, it should curve up as lag shortens the data. That’s not really an issue here though, it won’t vary noticeably in the central portion.)
W: “What your cited graph actually looks like, however, is that you forgot to detrend your datasets before doing the cross-correlation …”
The slope affects the correlation. If you want to know where the max correlation happens and what it’s value is, why would I want to distort both datasets by removing a spurious linear trend from both before doing CC?
Nir Shaviv says:
June 8, 2014 at 3:15 am
Nir, sorry for the confusion, I can see my exact wording wasn’t clear. I didn’t mean to ask for the actual records. All I need are the names of the stations that you used.
However, this is a perfect example of why transparency in the form of archiving the data as used is so important in science. It appears from your comments that you’ve forgotten that you didn’t use for your analysis the 177 stations chosen by Holgate. You mentioned the Holgate data in your study, but in your actual analysis, you used the 24 stations chosen by Douglas, so that’s what I need. As I said in my request,
All I need are the names of the 24 stations that were chosen by Douglas and used by you in your analysis. I can do the rest.
If you had archived your data and your code, you wouldn’t be bothered with having to go through these misunderstandings. In addition, we could examine your code and see if you’ve made any of the hundreds of common foolish mistakes that have bedeviled and tormented myself and other programmers since the invention of computers. I wrote my first computer program in 1963. It had bugs. Welcome to computing.
In addition, without your code it’s not clear what you’ve done. For example. Did you detrend the tidal records before you took the first differences? It makes a big difference, and you don’t mention it. Did you standardize the tidal records before you did the averaging? Again, it makes a big difference and you didn’t mention it.
Note that this is not a problem with your description. Your description is fine, neither better nor worse than the average for the breed. The problem is much deeper, and has nothing to do with you.
The problem is that English is far too vague and uncertain and imprecise a language to describe a series of computer operations. That’s why we don’t program in English—it is inadequate to the task. So no matter how detailed your English language description of your computer program might be, it’s not enough to answer all of the possible questions and reveal the hidden flaws.
Finally, even if you could describe precisely what it is that you think you did in the computer code, I can’t tell you how many times I’ve thought that I knew what I did in the code … and it turned out that the code was doing something completely different. So even if a scientist is perfectly candid and honest and detailed about what he thinks he did … in the real world, that means little about what the computer did when he wasn’t looking.
As a result, I fear that as it stands your study is not science at all. It is merely an advertisement for science. Science is transparent. Your work is totally opaque, and my motto is simple—no data, no code, no science. I can’t replicate your work without the data and code. I can’t identify hidden errors in your work without the data and code. I can’t find out if you detrended the tidal records without the data and code. In short, and sadly, as it stand your work is unfalsifiable because it cannot be examined.
As a result, it’s not science, it’s just an advertisement for your claims.
However, all is not lost, nothing is final. You could turn your study into real science by the simple expedient of archiving your data as used and your code as used so it can be examined for errors … as I and many other scientists do as a matter of course. It’s a pain, I hate cleaning up my code and putting the data in a useable form, but it only has to be done once for each study. Of course, I’m archiving data and code at the rate of a couple of studies a week, so archiving is more of an ongoing struggle for me than you, but hey … that’s science.
In any case, archive or not, it’s your choice. And in the meantime, a list of the 24 tidal stations that you used in your analysis would be much appreciated.
All the best, and again my thanks for you coming to defend your work, that’s science at its finest … or it would be if we could actually examine your work.
w.
Nir Shaviv says:
June 8, 2014 at 3:15 am
Nir, thanks for the reply. However, I fail to see how that applies here. YOU claimed that the HadISST dataset was relevant for showing a significant solar effect. I used YOUR choice of data and YOUR choice of methods (as best I understand them) and I found no such results.
How can that possibly be of no interest to you? I’ve shown that your claims are incorrect using your data and your methods, and your comment is “I couldn’t care less”? …
Say what? What am I missing here?
w.
Willis – Thanks for your reply to my last comment. I do accept that you have looked very diligently (and skilfully) at just about everything you can think of, and that the signal has not been there. I also accept that given your extensive efforts and others’ the probability of the signal existing is (to put it very conservatively!) small. Many thanks for trying, and many thanks for posting it all here. It’s part of what makes WUWT such an interesting and informative blog. [Thinks : if something that could be helpful to sceptics is demolished on a warmist site, there’s always the suspicion of bias. On WUWT it’s credible.]
Steven Mosher – I find your sneering comments illogical unjustified unworthy and unbecoming. Willis is looking for something, and he can’t find it, so others make suggestions as to where else he could look. So they are being helpful. Willis has looked in all the suggested places, and the thing he’s looking for isn’t in any of those places either. In all of that process, no-one has necessarily had any belief about whether the thing existed or not, an open mind is all that was needed. You are the one with the closed mind that is out of order.
I still think that Willis’ “Regular cycles which are apparently real appear, last for a half century or even a century, and then disappear for a century …” could be relevant. Does it necessarily mean that all such cycles are, in terms of what causes them, an illusion, or does it mean that when conditions change then effects change too? Even if effects disappear at times, wouldn’t they still be visible when averaged over the full record? The reality is that (a) I have to wait for someone to find the answer, or (b) Willis has found the answer (it’s an illusion), or (c) I have to find it myself. The last option, regrettably, is very unlikely.
Greg Goodman says:
June 8, 2014 at 4:16 am
——————————–
“My impression (which could obviously be mistaken) is that you failed to get any credible results or never even got as far as constructing the experiment.”
Greg, steady on. You are mistaken. I believe you may have mis-interpreted my comment about publishing results. I showed you not just a jpg of the revised experiment design, but a photograph of the very first of these experiments I built. This involved reflecting IR back to the surface of warm water. Results were published at Talkshop. My point about publishing results is this – one persons results carry far less weight than other persons replicating experiments.
I was not asking you to do the donkey work and run an experiment I had not run myself. Rather when you asked for proof, I gave you what I believed was the best proof possible, instruction on how to replicate the experiment.
I struggle to think of any more solid proof than an experiment you can replicate for yourself.
If you build the initial version of the experiment with IR reflected back to one cooling sample you will achieve ~1.5C divergence in ~30min in the evaporation constrained run. If you use a constant strong IR source as shown in the second version you will achieve over 5C of divergence. But I am not asking you to take my word for it. I am showing you how to check for yourself, just like Genghis is doing with IR thermometers and SST.
Nir Shaviv says:
June 8, 2014 at 3:15 am
Thanks for that, Nir. I don’t know if I folded the data “correctly” or not. It’s yet another example of the necessity for archiving your code. Your description is not entirely clear.
In your Figure 5 you are using an 11-year cycle. Unfortunately, there are a couple of ways to do that.
One way is the way I took. I set the start of the fold at the minimum, and I just graphed out the next 11 years. As you point out, with varying length cycles this is problematic towards the end.
It sounds from your description in this comment as though you might have done it the other way. This is to interpolate an e.g. 13-year cycle to fit it neatly into an 11-year series of bins. I don’t use this method myself, because it creates an artificial situation—not all of the cycles are the same length, but your method assumes that they are. You could use the same method to jam any disparage group of cycles of any length into a new cycle of any length … I just don’t see the theoretical justification for that procedure.
However, none of this is of import because your own figures do not show a statistically significant result. Only two out of the 11 averages of your folded data are statistically different from zero at a p-value of 0.05. Remember that at that p-value of 0.05, we expect to find one false positive every twenty trials. With 11 trials over the 11 years of your folded average, finding two such results is a totally unremarkable result. Finding two of them out of eleven has an overall p-value of 0.08, meaning your results are not statistically significant.
Now, you have stated above that you have used a chi-square test to show that your results are indeed significant, saying that you determined the significance …
However, the clearest mention of chi 2 in your paper is where you say you use a “sinusoidal least χ2 fit”. You also describe “A χ2 fitting …” and “The solid lines are χ2 fits to a harmonic variation”, and that’s it for chi-square. In other words, you only describe using a chi-square fit. Nowhere that I can find do you say anything about a chi-square significance test such as you describe above.
Let me say again that this is not your fault. The English language is simply not sufficiently precise and unambiguous to program computers, or to describe the action of the programs. That’s why we have computer languages … and it is also why it is so important that the code be archived, to avoid just these types of questions.
As a result, since your code is not archived I have no idea exactly how you “calculated the chi 2” or what you subsequently did with it. So I have to fall back on my own methods, which are to note that your result (only two out of the 11 of your folded averages are significantly different from zero at the p=0.05 level) is not at all unusual or statistically significant, it is expected … and it gets steadily worse when you use more than three solar cycles.
Finally, it appears that you do not understand the implications of subdividing the data and taking correlations with the parts. You point, for example, to the fact that the correlation is better with the Atlantic portion of the OHC data than with the Global data of which the Atlantic data is a part.
However, in general this is true for any dataset. If you split the dataset, one resulting subset will have a greater correlation with any given X than the other subset, they won’t be the same. You seem to think this is meaningful. And indeed, it may well be very meaningful … or not.
The part I can’t find any sign of you accounting for in your calculations is that now, instead of having one trial looking for a significant result, you have three trials (correlations with the full data, with the Atlantic data, and with the non-Atlantic data). And if you continue that process of subdivision, sooner or later you will find a result that is “significant” at say a p-value of 0.05 … so what? If you keep looking long enough for something with a one in twenty chance of a false positive, before long you will assuredly find it.
Now, the way folks deal with this issue is to notice that the more places you look, the more unusual your results need to be in order to be statistically significant. Finding a one-in-twenty result (p=0.05) is meaningless if you’ve conducted twenty trials, that’s what you’d expect to find.
A good approximation of the need for better and better results is that for statistical significance, you need to find a result that has a p-value which is equal to the single-trial p-value (typically 0.05 in climate science), divided by the number of trials. So in your case, with three trials, you need to come up with a result having a p-value of 0.05 / 3 ≈ 0.017 …
w.
PS—As I said in the head post, I would have handled the folding differently. Rather than use your method, which allows you to e.g. shoehorn a 13-year cycle equally easily into a 9-year or an 18-year interval, I would have aligned the data at the peaks, and then looked at six years before and six years after the peak year. Averaging that stack uses real data rather than involve creating pseudo-data by cramming a 13-year cycle into 11 years, and it minimizes the end effects.
Here is a comparison of the of Hadcrut3 and sun spot numbers from 1950 through 2014. There is a distinct 11 year cycle evident with both. spectrum
I used data from woodfortrees.org and analyzed it with Audacity.
The 11 year cycle is less evident when including older data. Maybe the older the data the more crap it is.
ralfellis says:
June 8, 2014 at 4:29 am
Thanks for that vote of confidence, Ralph. At times I think a stipend would be great … but then I realize that I’ll get slammed for my funding source, whatever it might be. Also, if I’m paid I’m obligated to write, and I don’t like that feeling. Even if the funder said “Just write when you want to”, it wouldn’t work—if I take the money, I feel obligated. This way, I’m free to write when and as I please, and pick and choose the subjects of interest … well, except when I get pulled back into the joy of the solar cycles, at least …
Regards,
w.
Crispin in Waterloo says:
June 8, 2014 at 9:19 am
———————————–
“Something else worth mentioning is that ocean water is quite reflective as well as having a high emissivity. I do not believe the story about water having an E of 0.7 – I have measured too much water with an IR thermometer to accept such a low number. The emissivity of water is in the high ’9′s’.”
This is getting a bit off topic, but I have been conducting some recent experiments into this issue. These involve measuring water surface temp with an IR thermometer under a cryo cooled “sky”. I can only get down to -40C at this stage. But with background IR minimised I need to adjust emissivity down to below 0.8 to get a reading matching surface thermocouple. An emissivity setting of 0.95 is fine for environmental measurement of water, but if that figure were used for calculating the radiative cooling rate of the oceans in the absence of atmosphere….
Genghis says:
June 8, 2014 at 4:31 am
Thanks, Genghis. Your claim is that the water just below the surface goes up and down in temperature … and the air just above the surface goes up and down in temperature … but the surface stays exactly the same, except for wind?
I fear that falls into the category of “extraordinary claims require extraordinary evidence”, particularly since scientific researchers find otherwise. There’s a good discussion of the issues in this PDF. Inter alia, they say
So they don’t find the result you mention. But take a look at the document, it is an in-depth discussion of the issues.
In addition, you should look at the UK Met Office discussion of the conversion of skin temperature to bulk temperature, which is here. As the other document said, the relationship between skin temperature and the lower layers is well-behaved enough to successfully model it. This is important, because the satellite surface temperature measuring is only measuring the skin temperature. So to be compatible with the normal SST, they use the calculations to convert from skin temperature (at a given time, place, and wind speed) to bulk temperature … again, well worth reading.
Ah, I understand now, said the long-time-married sailor …
w.
Greg Goodman says:
June 8, 2014 at 4:32 am
As I have said many times, the fact that the solar cycle is irregular does not prevent us from detecting it in the sun, using any one of a number of methods. As a result, the claim that the effect of the cycle is magically undetectable in climate datasets for unknown reasons won’t fly.
w.
Calculating for significance is as fraught with misguided creativity as calculating for linear trend lines. Trouble is, those who fail to understand the underlying math and proofs of data analysis think they can come up with a facsimile and call it good.
One of my favorites is a linear trend line through noisy data that was calculated by subtracting the first data point from the last data point and dividing by number of weeks between them to come up with the linear trend function. And then based on that result, use the function to substantially determine whether or not a student has a learning disability. When I protested, and stubbornly insisted that they should use linear calculations that were valid and reliable for noisy data instead of the made up of whole cloth calculation, I was threatened with a one day suspension.
Please folks, in this challenge by Willis to come up with something, don’t play around with statistical significance. Be conservative and use the gold standards.
RH says:
Here is a comparison of the of Hadcrut3 and sun spot numbers from 1950 through 2014. There is a distinct 11 year cycle evident with both. spectrum
I used data from woodfortrees.org and analyzed it with Audacity.
The 11 year cycle is less evident when including older data. Maybe the older the data the more crap it is.
====
I note in your “spectrum” link they both show 5.5y too. That is part of what makes up the shape of the solar cycle which rises quickly then tails off.
The late 20th c. period is one where it “works” which is why the question of cherry-picking arises. As you note earlier it works less well. It may have something to do with sampling biases in earlier data or just as likely the speculative “corrections” that are done to the data, mainly I think it is that the SST record quite a mix of cycles:
There is a fairly strong circa 9y component in most ocean basins. As this drifts in and out of phase with 11y cycles, as time progresses it will either add to or disrupt the 11y signal. This artificially increases it in much of the latter part of 20th c. and pretty much destroys it or pushes it out of phase in pre WWII period. Add to this that the “11y cycle” is a triplet of three close frequencies also interacting with each other and changing the profile and height of the solar peaks themselves plus a smaller 22y component.
As far as I have been able to tell, the 9y signal is similar to but a bit larger than the solar signal. About half of what appears to be solar correlation when it “works” is in-phase contribution of 9y.
The “9y” cycle seems more stable than the complex solar signal, it appears to be 9.05 to 9.1y , with various authors giving various margins of uncertainty, I suspect it is quite close to that central value.
All this means that tracking down any match to the solar signal is not going to just jump out of the page as some seem to expect it will. “It’s the sun stupid” is well, stupid.
IPCC seem to favour 0.1K pk-pk variation in surface temps over the 11 year cycle. It may be a fair bit stronger but that is order of magnitude concerned. We are looking for that against a record with annual swings about two orders larger with opposing phases in asymmetric hemispherical variations and 6mo tropical seasons. Plus all the rest of the churning climate system.
Significance levels judges against naively simplistic “red noise” models are of little relevance.
The school of AGW says it’s GHG+stochastic , in that ‘one variable’ model and under the assumption the rest is chaotic, the red noise test makes sense. Under a model that anticipates solar, lunuar, anthro and other ( possibly driven ) factors + noise , the individual components will be small and will not be “significant” against a simplistic statistical model where everything else is red.
Neither is unconstrained red noise a suitable model for variation in variables that are not free to do a ‘drunkards walk’ across the park but are instead bounded by negative feedbacks to remain on the path.
I think all that answers some of Mike Jonas’ points too.
W: “As I have said many times, the fact that the solar cycle is irregular does not prevent us from detecting it in the sun, using any one of a number of methods. ”
Agreed, my point was that is it not a simple fixed 11 years.
W: “As a result, the claim that the effect of the cycle is magically undetectable in climate datasets for unknown reasons won’t fly.”
Is that supposed to relate to my quote? I did not say it was, I said it has be sought as directly to the solar signal , because of its irregular nature.
How are you getting on with interpreting the 0.25 correlation coefficients? What does your ‘program’ give for circa 1700 data points?
If you are having trouble with my cross-correlation, do your own with the full monthly data, without averaging, “binning”, detrending, just a straight CC. Correlation is simply and uniquely defined and calculable at each lag value.
Do you still think it is necessary to detrend before doing CC ? That would seem to be an error to me in the context of assessing magnitude and timing of peak CC.