Readers may find the title familiar, that’s because Basil Copeland and I also did a paper looking at solar signatures in climatic data, which has received a lot of criticism because we made an analytical error in our attempt. But errors are useful, teachable moments, even if they are embarrassing, and our second attempt though, titled,
hasn’t been significantly challenged yet that I am aware of. Basil and I welcome any comments or suggestions on that work.
In our work, we used Hodrick-Prescott filtering to extract the solar cycle signal from the HadCRUT temperature dataset. In this paper the data are extracted from the ECA&ECD database (available via http://eca.knmi.nl ). According to the paper, they are “using a nonlinear technique of analysis developed for time series whose complexity arises from interactions between different sources over different time scales”. Read more about it in the paper. In both our paper, and in this one, a solar signature is evident in the temperature data. – Anthony
Evidence for a solar signature in 20th-century temperature
By Jean-Louis Le Mouel, Vincent Courtillot, Elena Blanter, Mikhail Shnirman (PDF available here)
J.-L. Le Mouël et al., Evidence for a solar signature in 20th-century temperature data from the USA and Europe, C. R. Geoscience (2008), doi:10.1016/j.crte.2008.06.001
Abstract
We analyze temperature data from meteorological stations in the USA (six climatic regions, 153 stations), Europe (44 stations, considered as one climatic region) and Australia (preliminary, five stations). We select stations with long, homogeneous series of daily minimum temperatures (covering most of the 20th century, with few or no gaps).We find that station data are well correlated over distances in the order of a thousand kilometres. When an average is calculated for each climatic region, we find well characterized mean curves with strong variability in the 3–15-year period range and a superimposed decadal to centennial (or ‘secular’) trend consisting of a small number of linear segments separated by rather sharp changes in slope.
Our overall curve for the USA rises sharply from 1910 to 1940, then decreases until 1980 and rises sharply again since then. The minima around 1920 and 1980 have similar values, and so do the maxima around 1935 and 2000; the range between minima and maxima is 1.3 °C. The European mean curve is quite different, and can be described as a step-like function with zero slope and a ~1 8°C jump occurring in less than two years around 1987. Also notable is a strong (cold) minimum in 1940. Both the USA and the European mean curves are rather different from the corresponding curves illustrated in the 2007 IPCC report.We then estimate the long-term behaviour of the higher frequencies (disturbances) of the temperature series by calculating the mean-squared interannual variations or the ‘lifetime’ (i.e. the mean duration of temperature disturbances) of the data series.We find that the resulting curves correlate remarkably well at the longer periods, within and between regions. The secular trend of all of these curves is similar (an S-shaped pattern), with a rise from 1900 to 1950, a decrease from 1950 to 1975, and a subsequent (small) increase. This trend is the same as that found for a number of solar indices, such as sunspot number or magnetic field components in any observatory. We conclude that significant solar forcing is present in temperature disturbances in the areas we analyzed and conjecture that this should be a global feature.
…
We find that station data are well correlated over distances in the order of a thousand kilometres. When an average is calculated for each climatic region, we find well characterized mean curves with strong variability in the 3-15-year period range and a superimposed decadal to centennial or ‘secular’ trend consisting of a small number of linear segments separated by rather sharp changes in slope. Our overall curve for the USA rises sharply from 1910 to 1940, then decreases until 1980 and rises sharply again since then. The minima around 1920 and 1980 have similar values, and so do the maxima around 1935 and 2000; the range between minima and maxima is 1.38C. The European mean curve is quite different, and can be described as a step-like function with zero slope and a 1.8C jump occurring in less than two years around 1987. Also notable is a strong (cold) minimum in 1940. Both the USA and the European mean curves are rather different from the corresponding curves illustrated in the 2007 IPCC report.
…
We then estimate the long-term behaviour of the higher frequencies (disturbances) of the temperature series by calculating the mean-squared interannual variations or the ‘lifetime’ (i.e. the mean duration of temperature disturbances) of the data series. We find that the resulting curves correlate remarkably well at the longer periods, within and between regions. The secular trend of all of these curves is similar (an S-shaped pattern), with a rise from 1900 to 1950, a decrease from 1950 to 1975, and a subsequent (small) increase. This trend is the same as that found for a number of solar indices, such as sunspot number or magnetic field components in any observatory.
…
We conclude that significant solar forcing is present in temperature disturbances in the areas we analyzed and conjecture that this should be a global feature.
We have also shown that solar activity, as characterized by the mean-squared daily variation of a geomagnetic component (but equally by sunspot numbers or sunspot surface) modulates major features of climate. And this modulation is strong, much stronger than the one per mil variation in total solar irradiance in the 1- to 11-year range: the interannual variation, which does amount to energy content, varies by a factor of two in Europe, the USA and Australia. This result could well be valid at the full continental scale if not worldwide. We have calculated the evolution of temperature disturbances, using either the mean-squared annual variation or the lifetime. When 22-year averaged variations are compared, the same features emerge, particularly a characteristic centennial trend (an S-shaped curve) consisting of a rise from 1920 to 1950, a decrease from 1950 to 1975 and a rise since. A very similar trend is found for solar indices. Both these longer-term variations, and decadal and sub-decadal, well-correlated features in lifetime result from the persistence of higher frequency phenomena that appear to be influenced by the Sun. The present preliminary study of course needs confirmation by including regions that have not yet been analyzed.
Another worry about this paper… I can’t reproduce their solar magnetic curve from a naive running mean of sunspot numbers, which I think I would expect to be able to (should they be correlated, Leif?).
http://www.woodfortrees.org/plot/sidc-ssn/from:1920/to:1990/mean:120
The 1955-1960 peak is there, as is the little bounce around 1970, but their curve is missing the marked rise around 1980, and without that, their correlations with temperature wouldn’t be anywhere near as compelling. It’s as if their filtering method (if that’s what it is?) has a high-pass at around 20 years.
Also, it’s pretty clear to me that the sunspot peak significantly lags the temperature peak – by about 2 decades!:
http://www.woodfortrees.org/plot/sidc-ssn/from:1920/to:1990/mean:120/normalise/plot/hadcrut3vgl/from:1920/to:1990/mean:120/normalise
So I’m now seriously wondering whether I understand what it is they claim to be plotting… I’m going to have to read up about this “lifetime” business!
(oh, my posts have gone out of order because the first one had 3 links and probably ended up in the moderators spam bucket – sorry)
woodfortrees (Paul Clark) (01:45:46) :
Another worry about this paper… I can’t reproduce their solar magnetic curve from a naive running mean of sunspot numbers, which I think I would expect to be able to (should they be correlated, Leif?).
Yes, one would think so. But they use the geomagnetic variations at just one station [ESK in Northern UK, which BTW had data problems] as a proxy for solar activity, which is strange because we have solar data, so why use a troubled proxy? That was another of the [many] reasons I recommended rejection of the paper.
There is a problem with the peer review system: Journals are rated [the ‘impact factor’] and scientists want to have their papers published in the highest rated journal in their field, so often they submit the paper to the highest rated journal first. Now, one reason a journal has a high impact factor is its rigorous review process, so they also have the highest rejection rate. If a paper is rejected, then you submit to the next highest journal, after rejection there, you submit to the next, etc, until finally you get it accepted in minor journal, either because they found a friendly reviewer or an incompetent one, either reason not being satisfactory from the viewpoint of the public that relies on the review process to place trust in the paper. One solution to this would be to require that the reviews [all of them, from the first journal to the last that finally accepted the paper] be public and be [electronically] attached to the paper. We are not there yet, but it would be a great help in judging the trustworthiness of the paper.
Basil
Above I asked a few questions I’ll repeat them here with a few corrections/additions since you have not answered them.
From Your posting
http://wattsupwiththat.com/2009/05/23/evidence-of-a-lunisolar-influence-on-decadal-and-bidecadal-oscillations-in-globally-averaged-temperature-trends/
Perhaps you can explain how fig 5 compares to fig 2. Fig 5 shows a sudden drop at >56 years and more detail than any FFTs can show at these periods despite the same data being used?
To me it looks as if a band pass filter has been used with a passband of 8 to 22 year period. Is this true?
The HP (not high pass!) filter is good and simple for smoothing data but I know nothing of its internal workings. I would therefore hesitate to use it not knowing how it mungs the data! For example this from wiki:
Drawbacks to H-P filter
The Hodrick-Prescott filter will only be optimal when:[5]
Data exists in a I(2) trend.
If one-time permanent shocks or split growth rates occur, the filter will generate shifts in the trend that do not actually exist.
Noise in data is approximately Normal~(0,σ²)(White Noise).
Analysis is purely historical and static (closed domain). The filter has misleading predictive outcome when used dynamically since the algorithm changes (during iteration for minimization) the past state (unlike a moving average) of the time series to adjust for the current state regardless of the size of lambda used
(I checked it against a moving average and similar outputs could be obtained)
Is this filter safe to use when processing this sort of data?
The MTM FFT software seems to put little hats in black on the trace – are these peaks actually present at these amplitudes shown or is this just the software pointing out where peaks could be if they were harmonically related?
In my plots a FFT of the TSI/SSN was included – Is is not a single noise free spike but one would expect to see this shape in the temperature plots if it was there. TSI shows no 22 year peak .
http://img21.imageshack.us/img21/9826/tsifft.jpg
Please ignore the CET plots as you seem to find them flawed.
From another discussion on wuwt the conclusion was drawn that it was not possible to get a high frequency to “pump” a subharmonic (11 year will not cause a 22 year cycle) The other way is OK (22 year cycle can cause a 11 year harmonic) (I simulated it using tuned circuits in a simulation package)
PS I think the tone of your reply to me was a bit below the belt and unecessary
Basil (09:31:06) :
And I’ve responded just as many times. …Rather than just trumpet your fft charts…
Someone asked a question and I replied – that is all. No trumpeting. And you certainly have not pointed out where my (or for that matter Leif’s) plots of FFTs are invalid!
Sometimes, I think that this “Sun Is Responsible” argument is silly. Everyone knows that the Earth would not be warm, or as warm/hot as it is, without dear Sun. Yes, when the sun comes up, it is warmer on one’s spot on Earth — but it is cooler on the opposite side. Yes, when sun spots have “disappeared”, the Earth seems to have cooled considerably, but so far not predictably so. What it seems we are interested in (or at least I am and why I am choosing WUWT over most other blogs) is how the Sun interacts with Earth’s environment and position in the galaxy/universe. I am also interested in the chaotic nature of Earth’s weather/climate (or at least it seems chaotic at present).
I see no problem with one scientist saying that a perspective has been proven irrelevant — as-long-as-the-evidence-is-presented. And I see no problem with the scientist of the forcefully disputed position returning with more evidence in support of his/her ideas. Let the disputation focus on the hypothesis and evidence. I have experienced some very productive “yelling” on your blog, Anthony. Scientific disputations are not business meetings; in that regard I disagree with you. And yet it is yours, for which I am very grateful. Let the good times roll.
Just for interest some time ago I compered Hoderik Prescott, Eponential smoothing, moving average, and bandpass filtering on hadcrut data:
http://img14.imageshack.us/img14/1282/hodrickprescottfilter.jpg
HP filtering on monthly data with a factor of 20000 gave a similar response to 3 year moving average with a few differences – some ripples removed, some peaks shifted and end of data handled differently (on the moving average there should be no valid results within 1.5years of the data ends.
bill (06:31:40) :
since you have not answered them.
you certainly have not pointed out where my (or for that matter Leif’s) plots of FFTs are invalid!
Well I pointed out that although the average cycle length is 11.01 years, there is almost never an 11 year solar cycle. So why expect temperature plots to conform under FFT? Especially with the kind of biennial swings caused by other factors we see.
If you want to see the solar signal in the temperature series just smooth the temp series on a 43 month moving average like this:
http://www.woodfortrees.org/plot/hadsst2gl/from:1970/to:2009/mean:43/detrend:0.5/plot/sidc-ssn/from:1970/to:2009/scale:0.001
I don’t know why I’m bothering, because I replied to a few of your posts now, and you never return the compliment.
And I forgot to thank Paul Clark for the use of his excellent graphing facilities.
Thanks again Paul.
tallbloke (08:32:30) :
If you want to see the solar signal in the temperature series just smooth the temp series on a 43 month moving average like this
Even if the period changes a bit there would still be an FFT signal, the peak would just be broader. And the signal in your graph disappears when you just go a bit further back, e.g. to 1880. And finally, the ‘signal’ is as small as the 0.1 degree commonly predicted anyway.
Or if you like playing with FFT’s over longer timescales:
http://www.woodfortrees.org/plot/hadsst2gl/from:1910/to:2009/mean:43/detrend:0.5/fourier/low-pass:10/inverse-fourier/normalise/plot/sidc-ssn/from:1910/to:2009/scale:0.0005/fourier/low-pass:10/inverse-fourier/normalise
Can you see the solar signal in the data now?
This shows it even better. Notice the peak around the strong el nino phase in the 1930’s followed by the slump and late response to the following cycle. The oceans have a 60 year cycle, guess what’s coming soon to a skating rink near you.
http://www.woodfortrees.org/plot/hadsst2gl/from:1910/to:2009/mean:43/detrend:0.5/fourier/low-pass:12/inverse-fourier/normalise/plot/sidc-ssn/from:1910/to:2009/scale:0.0005/fourier/low-pass:12/inverse-fourier/normalise
tallbloke (08:32:30) :
I didnot know you were expecting an answer.
Taking your link and correcting it
to remove the de-trend (why remove the trend?)
include a few more years (What’s wrong with the earlier data?)
Take the mean over a whole 2 years (43 months – why? )
gets you this (woodfortrees is a wonderful cherry orchard!):
http://www.woodfortrees.org/plot/hadsst2gl/from:1900/to:2009/mean:24/plot/sidc-ssn/from:1900/to:2009/scale:0.001
Doesn’t look so good now, does it?
The FFTs I made all had a plot of the TSI or SSN FFT included so the pattern of the spectrum of SSN was known.
This pattern does not show up above the noise. Other signals do, but these are not sun related. If it’s in the noise it will have no effect – even if by filtering and other DSP techniques you can show that it is actually present (I know it must be present because variations in TSI is a fact).
bill (06:31:40):
Today, I only have time for a quick note, namely, that certain nonlinear systems can indeed “pump” subharmonics of the forcing. Linear circuits are incapable of subharmonic response.
P. S. Have you ever plotted your FFT results as an equi-spaced sequence of frequencies at which they are computed. It is in the frequency domain–and not that of the period you plot–that the Parseval relationship between the magnitudes of the F. coefficients and the total variance applies. You might also be surprised how much those coefficients vary with slight changes in record length. That’s why raw FFT periodograms are not relied upon in power spectrum estimation.
bill (09:06:23) :
Taking your link and correcting it
to remove the de-trend (why remove the trend?)
include a few more years (What’s wrong with the earlier data?)
Take the mean over a whole 2 years (43 months – why? )
I remove the trend to isolate the solar signal from oceanic cycles as far as possible.
1910 on is more reliable SST data IMO.
43 months because that is 1/3 of the solar cycle. (2 years -why?)
Anyway, none of that matters, because I now understand that you simply don’t want to see the solar signal, and will do anything necessary to make it go away. Which is fine, we now know where each other stands, and I won’t waste anymore of your time. Or mine.
Oh, last thing, you didn’t comment on my FFT’s, but here’s the last one, with the earlir data included. You can see the descent from the previous 60 year peak on this one, and join the dots, if you can bring yourself to make a useful interpretation.
http://www.woodfortrees.org/plot/hadsst2gl/from:1880/to:2009/mean:43/detrend:0.5/fourier/low-pass:15/inverse-fourier/normalise/plot/sidc-ssn/from:1880/to:2009/scale:0.0005/fourier/low-pass:15/inverse-fourier/normalise
John S. (09:21:07) :
You might also be surprised how much those coefficients vary with slight changes in record length.
This effect can be clearly seen in this raw FFT plot of the sunspot series 1700-2008:
http://www.leif.org/research/FFT-Power-Spectrum-SSN-1700-2008.png
The FFT was calculated for 1700-2008, then for 1701-2008, 1702-2008, etc, a total of 17 curves. The length-effect, of course, becomes greater the longer the period is. On the other hand the significant peaks are still there, no matter what the record length is.
Leif Svalgaard (08:48:45) :
tallbloke (08:32:30) :
If you want to see the solar signal in the temperature series just smooth the temp series on a 43 month moving average like this
Even if the period changes a bit there would still be an FFT signal, the peak would just be broader. And the signal in your graph disappears when you just go a bit further back, e.g. to 1880. And finally, the ’signal’ is as small as the 0.1 degree commonly predicted anyway.
Well as you know Leif, there are other things affecting the temperature series apart from TSI. However from the ‘big signal’ helped along no doubt by El nino dominted phases of around 0.6C to the ‘little signal’ of around 0.17C dampened by La Nina phases, there would seem to be an average around 0.38C, which is half a centuries worth of ‘global warming’ is it not? Probably more once you take Anthony’s work into account.
So, by no means a minor 0.1C blip, and when you also take into account the cumulative running total method I developed, it seems the sun is a much bigger player in the climate game than youtry to make out. And that’s without having to invoke geomagnetism, extra UV or anything else.
tallbloke (09:24:51) :
but here’s the last one, with the earlier data included.
As the peaks sometimes are in phase and sometimes not, all you show is that there are decadal-type variations, which nobody disputes. They just don’t line up with the solar cycles very well, which is the important point. And there should be a 0.07 degree solar signal. An interesting thing for you to do, would be to take the SSN, multiply it by 0.07 and divide by 115 [the average SSN peak value], then subtract that from the temperature curve and look again.
My mistake, messed up my scaling. 0.2C to 0.1C. Still not insignificant when heat builds up in the ocean over periods of high solar activity.
tallbloke (09:41:48) :
Still not insignificant when heat builds up in the ocean over periods of high solar activity.
If you heat something up [the oceans] they radiate more at the higher temperature. Heat does not build up. If you suddenly increased TSI by 0.1% the new temperature would be 0.07 degrees higher no matter for how long the increase lasts.
Leif Svalgaard (10:22:04) :
tallbloke (09:41:48) :
“Still not insignificant when heat builds up in the ocean over periods of high solar activity.”
If you heat something up [the oceans] they radiate more at the higher temperature. Heat does not build up. If you suddenly increased TSI by 0.1% the new temperature would be 0.07 degrees higher no matter for how long the increase lasts.
Perhaps I should add that it might take the oceans a while to heat up the 0.07C [and a while to cool off again, if the TSI increase went away], but you don’t get it any higher than the 0.07C increase. And any delay in heating the oceans would destroy the phasing, wouldn’t it? [unless you postulate that the delay is precise a whole multiple of solar cycles].
To me it looks tantalisingly like there is a link between Earth climate and solar activity which is greater than can be explained by changes in TSI alone.
If we better understood the mechanisms which lie behind the suns 22y cycle, this would perhaps give us a better understanding of what we are missing.
My thinking is that we could, perhaps, make headway by looking for something that could be effecting the chaotic processes of both the sun and the earth. An external force could be effecting the whole of the solar system on a periodic basis, and have slightly differing effects on both the earth and sun regarding timing of the manifestation of event and the effects on differing mechanisms.
It sometimes only takes a small nudge to a chaotic system to cause a large event.
Hi Leif, thanks for your considered reply. The thermal expansion of the ocean that has to have taken place to account for the observed sea level rise demands that the heat is permeating down a long way below the depth commonly asserted as the ‘mixing layer’ (Levitus et al).
The atmosphere limits the amount of heat which can escape from the ocean the surface reaches a max of around 29C. Solar energy penetrating the ocean and warming the surface layer beyond that maximum causes the excess heat to propagate downwards. Where else can it go?
0.07 degrees is pretty close to the decadal warming rate so it seems to me that any residual heat heading down into the lower strata towards the thermocline could indeed exhibit a decadal oscillation as it rises back to the surface and reinforces or or otherwise modulates the current solar cycle. There is as much thermal capacity in the top 2.5m of the ocean as there is in the entire atmosphere above it, and the heating goes down to a thousand metres or more. All that heat can’t get out quickly, it has to wait for when the atmosphere is cooler during lower points in solar cycles or runs of lower solar cycles.
The ‘out of phase’ state the ocean gets into is also being caused during big cooling or warming periods as you can see from the FFT’s. This is when the SST has ‘overshot’ during a run of el nino’s, then crashes as in the 1940’s and now. The rebound keeps the temperature dropping as the next solar cycle is rising, and the ocean has to ‘play catch up’ to the higher solar signal during the following cycle max. As the temperature of the ocean heads up every 60 years, it almost smoothes away the solar signal, but you can see it is still there if you change the low pass filtering rate on the calcs.
I remember a year ago on CA when you said to me you thought there might be a 10 year ‘lag’ in the system. What prompted you to say that if no heat is stored? No-one has a satisfactory explanation for the 60 year cycle as far as I know, that’s why I’m pushing ideas around it. I’m not trying to sell you a cut and dried theory, and thanks again for your insight which informs me and makes me think on my feet.
By the way Leif, interesting peaks in your SSN power spectrum at 50 and 100 years.
🙂
Correction:
residual heat heading down into the lower strata towards the thermocline could indeed exhibit a decadal oscillation as it rises back to the surface and reinforces or or otherwise modulates the ocean’s response to the current solar cycle.
tallbloke (11:43:59) :
thanks again for your insight which informs me and makes me think on my feet.
Taking advantage of the above: I think you are confusing storage and temperature. If I increase the input (TSI) abruptly [as happens every 11 years – abruptly because the rise time is short], the upper layer [and SST and temps in general] will heat up fast [matter of weeks]. If I keep the heat on, the upper layers will not heat to any higher temperature, but the heat will penetrate deeper and deeper into the ocean [or the Earth (boreholes) for that matter]. Conversely, when you down down the heat, the heat already stored in the system will keep the temperature at the higher level for some time.
tallbloke (11:57:13) :
By the way Leif, interesting peaks in your SSN power spectrum at 50 and 100 years.
Whether they are really separate or just an expression of the fact that there is power in a broad region from 40 to 110 years [perhaps centered on the ~88 year Gleissberg cycle] the sunspot record is really too short to tell.