Scientists find errors in hypothesis linking solar flares to global temperature
From Physorg.com. h/t to Leif Svalgaard who offers this PDF with this diagram that makes it all clear.
In contrast to a previous analysis, a new study has shown that the distributions of (a) the global temperature anomaly by month since 1880 and (b) the solar flare index by day over a few solar cycles are fundamentally different. One feature the detrended data do have in common is self-similarity: the probability density functions are the same on different time scales, which means that neither can be described as Lévy walks. Image credit: Rypdal and Rypdal.
(PhysOrg.com) — The field of climate science is nothing if not complex, where a host of variables interact with each other in intricate ways to produce various changes. Just like any other area of science, climate science is far from being fully understood. As an example, a new study has discredited a previous hypothesis suggesting the existence of a link between solar flares and changes in the earth’s global temperature. The new study points out a few errors in the previous analysis, and concludes that the solar and climate records have very different properties that do not support the hypothesis of a sun-climate complexity linking.
In a handful of studies published in Physical Review Letters between 2003 and 2008, a team from Duke University and the Army Research Office including Nicola Scafetta and Bruce West analyzed data that appeared to show that solar flares have a significant influence on global temperature. Solar flares, which are large explosions in the sun’s atmosphere that are powered by magnetic energy, vary in time from a few per month to several per day. Although solar flares occur near sunspots, their frequency variation occurs on a much shorter time scale than the 11-year sunspot cycle. In their studies, the researchers’ results seemed to show that data from solar flare activity correlates with changes in the global temperature on a short time scale. Specifically, their analysis showed that the two time records can both be characterized by the same Lévy walk process.
However, in the new study, which is also published in Physical Review Letters, Martin Rypdal and Kristoffer Rypdal of the University of Tromso in Norway have reexamined the data and the previous analysis and noticed some shortcomings. One of the biggest causes of concern is that the previous analysis did not account for larger trends in factors that affect solar flares and global temperature. For instance, the solar cycle has its 11-year periodic trend, where periods of lots of sunspots cause larger numbers of solar flares. Likewise, the global temperature anomaly has numerous other factors (a “multi-decadal, polynomial trend”) that impacts global temperature fluctuations. By not detrending this data, the analysis resulted in abnormally high values of certain variables that pointed to Lévy walk processes. By estimating the untrended data, Rypdal and Rypdal hypothesized that the solar flare records might be described by a Lévy flight, while the global temperature anomaly might obey a distribution called persistent fractional Brownian motion.
Read the entire article here at Physorg.com
A preprint of the paper is available here
Practice making your own Levy walks here
Leif Svalgaard (07:21:34) : I accused you of cherry picking, not the authors
You accused the authors of cherry-picking (you accused me afterwards, when I asked you to back up your claim, which you still haven’t done) :
jinki (16:36:39) : Two proxy records in the Kirkby paper in general agreement, that’s all any reasonable person could expect.
Leif Svalgaard (22:35:05) : No, these are extraordinary claims, and require extraordinary evidence, not just cherry picked weak correlations.
Now you admit that the papers admit to conflicting results.
I never denied it?? Some conflicts arise from the a lack of understanding, by the way.
Now, instead of just reviewing papers or reviewing reviews of papers that review yet other papers, there are scientists out there [ourselves included] that are actually trying to do science and advance the field.
Examing the literature and ‘doing science’ are not mutually exclusive.
Jasper Kirkby, the author of ‘Cosmic Rays and Climate’ review, also does experimental work.
johnythelowery (04:58:23) :
“Is the earth’s atmosphere acting like a eye-ball and lensing and focusing a .1K input at the highest level to a 1K affect in a very localised band according to the density of the air in the various levels of Wilde’s atmosphere?”
Hi John, there’s no known lensing of the photonic solar radiation.
oneuniverse (09:30:44) :
You accused the authors of cherry-picking (you accused me afterwards, when I asked you to back up your claim, which you still haven’t done)
All authors cherry pick in this field [as far as I can see]. They use the Group Sunspot number when that supports their claim, they use Hoyt&Schatten TSI when that supports their claim, they use the 10Be core that best support their claim, the use the doubling of sun’s magnetic field, if that supports their claim, they use cosmic ray records that show the largest modulations, if that supports their claim, they use polar region influx instead of the important equatorial one, if that supports their claim, etc. If you claim to know the literature you’d know this.
Some conflicts arise from the a lack of understanding, by the way.
And you think that lack of understanding on part of the author of a review paper lends credence to it?
oneuniverse (09:30:44) :
jinki (16:36:39) : Two proxy records in the Kirkby paper in general agreement, that’s all any reasonable person could expect.
At the ISSI workshop I referred to on the Lockwood-thread http://www.issibern.ch/workshops/cosmicrays/ a paper by Joos Fortunat gives a ‘paleo-perspective on the carbon cycle – climate system:
[also here: http://www.leif.org/research/Joos-Bern-2010.ppt ].
“Conclusion: contribution to 20th century warming is less than 0.15K for all solar scalings” including, of course, solar modulation of 14C.
As long as we are down in that range, I have no reason to object, as that is what observationally driven modelling and theory show we should have.
May I suggest that further discussion on CRs be taken to another, more appropriate thread. This one here is for Levy-flights and S&W. Sadly, thread-hijacking is so prevalent, let’s try to decrease it a bit.
Leif Svalgaard (21:51:10) :
“The mistake is to try to model SFI as a Levy-walk.”
we explicitly excluded SFI, we are not using that record in any way. It is R&R that wanted to use it claiming that they did the same thing that we do.
Nicola Scafetta (15:02:16) :
we explicitly excluded SFI, we are not using that record in any way.
Help me out here. Direct me to the paper where you say “we explicitly exclude SFI”. I don’t remember seeing that. On the other hand, while you often explain in great detail where you get the temperature data from, many of your papers talk about solar flares being used, without specifying where you get the data from. Perhaps elaborate on that a bit.
Leif Svalgaard (18:47:25)
The Scafetta & West paper under discussion cites the following paper for the analysis of the solar flare data :
“Diffusion entropy and waiting time statistics of hard-x-ray solar flares”, P. Grigolini, D. Leddon, N. Scafetta 2002
“The data are a set of 7212 hard x-ray peak flaring event times obtained from the BATSE/CGRO (Burst and Transient Source Experiment aboard the Compton Gamma Ray observatory
satellite) solar flare catalog list. The data is a 9-year series of events from 1991 to 2000.”
That covers the entire period the satellite was in use. This data is not the SFI, which is a synthetic index maintained at Bogazici University .
However, I’m not sure where the figure of 7212 solar events is coming from – the summary of BATSE triggers shows 1192 solar events out of 8021 triggers (a more detailed list at ftp://umbra.nascom.nasa.gov/pub/batse/batse0/burst_trigger.lst ).
oneuniverse (21:54:57) :
The Scafetta & West paper under discussion […] solar flare data :
“The data are a set of 7212 hard x-ray peak flaring event times obtained from the BATSE/CGRO (Burst and Transient Source
Thanks for that info. It might be interesting to see R&R’s response. On the other hand, the data period is rather short, only 9 years. The main argument of R&R, as far as I can tell, is that the PDF for the temperature does not have the sharp peaked structure of a Levy distribution.
In my opinion just because two distributions look the same [which they btw don’t] it is not shown that they are caused by the same physical process. Perhaps it would work the other way: if the distributions are different, that would be an argument against a similar physical processes. It would be interesting if S&W would actually plot the two distributions for us all to see.
oneuniverse (21:54:57) :
The Scafetta & West paper under discussion […] solar flare data :
“The data are a set of 7212 hard x-ray peak flaring event times from the BATSE/CGRO”
Colleagues of mine have publish a ‘Science Nugget’ on the waiting time distribution for flares:
http://sprg.ssl.berkeley.edu/~tohban/wiki/index.php/Waiting_Times_of_Solar_Hard_X-Ray_Flares
They find that “statistical distributions of waiting times observed during three solar cycles with different instruments are fully consistent with a nonstationary Poisson process; this corroborates the conclusion that solar flares are indeed a phenomenon with self-organized criticality. http://en.wikipedia.org/wiki/Self-organized_criticality
Such processes are scale-invariant, as are Levy-flights, but Levy-walks are not scale-invariant [or self-similar].
Leif Svalgaard (22:41:56) :
Colleagues of mine have publish a ‘Science Nugget’
The nugget is based on this preprint:
http://arxiv.org/PS_cache/arxiv/pdf/1002/1002.4869v1.pdf
Leif Svalgaard (22:45:56) :
Colleagues of mine have publish a ‘Science Nugget’ […]
The nugget is based on this preprint:
http://arxiv.org/PS_cache/arxiv/pdf/1002/1002.4869v1.pdf
One of the conclusions is:
“1. Waiting time statistics gathered over a relatively small range of waiting times, e.g., < ∼ 2 decades as published by Pearce et al. (1993) or Crosby (1996), does not provide sufficient information to reveal the true functional form of the waiting time distribution.”
This would then also seem to be a problem with the 9-yr series used by S&W.
I follow the discussions here with great interest. I will comment on some of the stuff chronologically as I read – on points where I believe I can contribute constructively. I will start with a remark made on April 12., at 10.00.30. Here Tamara writes:
“The paper may in fact refute Scafetta and West, but refuting this hypothesis doesn’t generate evidence for AGW. In fact, I would think that showing that the GTA exhibits “persistent fractional Brownian motion” would suggest the opposite. Brownian increments are supposed to be random, independent, and equally likely to occur in either direction. How does this correlate with constantly increasing GHG concentration?”
I totally agree that falsiying S&W does not verify AGW. Those are not our words, but the journalist’s. However, while it is true that Brownian increments are random, FRACTIONAL Brownian increments are not. The GTA time-series exhibits a long-range memory, where the strength of this memory is characterized by the Hurst exponent H. However, one of the main points in our paper is that the strong multidecadal trend observed in the GTA signal is NOT an inherent part of the fractional Brownian motion (fBm). In fact if you produce a synthetic fBm signal and fit a fourth order polynomial to define such at trend, you will always find a much weaker trend than we have in the real GTA. This is why we have to subtract this trend from the GTA before we analyze it to find the properties of the random component of the signal. When we do this we find that it is an fBm with H=0.65. If there is an influence of the increasing GHG concentration on the GTA signal, the information about this is in the slow trend component, not in the fBm component of the signal. The memory in the fBm component, which we observe on time scales from a month up to a about ten years, we believe is due to internal dynamics in the climate system.
April 12 (10.16.39) bryan writes:
“The two graphs might indicate more similarity if they were displayed in the same timeframe. but the Solat Flare index is displayed at 15000 days and the Global temp is displayed at 45000 days (figuring a rough 30 days per month).”
S&W analyze catalogs of flare onsets, and make statistics for waiting times between them. They find some memory in that statistics for time scales up to 3 months, but not longer. Our analysis of the SFI, shows that it is adequately modeled as a Levy flight, which is a process without memory. We have also defined flare onsets from the SFI by counting the times when the index grows above a certain threshold, and find that there is no memory on time scales longer than a month when the quasi-periodic trend due to the solar cycle has been removed. However, our analysis of the GTA signal shows long-renge memory on scales up to 100 months.
A natural question to ask is: is there any reason to conjecture that the long-range memory in the GTA that lasts up to a decade has its origin in the sun when there is no such memory in the solar flare data?
On April 12 (11.08.41) James F. Evans write:
“While it’s admirable that the scientists are so up-front with their goals and purposes, this kind of pointed “outcome” oriented agenda should make readers cautious when considering what weight to give the conclusions of this paper and other papers who’s authors state similar “outcome” oriented agendas.”
I think James misunderstands a bit. Our “outcome oriented agenda” is to falsify hypotheses in general, not only those that are incompatible with the AGW theory. This agenda is consistent with Carl Popper’s philosophy that scientific theories can be falsified, but never verified. We would be very proud of ourselves if we could falsify the AGW theory, for instance by proving that increased consentration of greenhouse gases is not the dominant cause of recent global warming.
Unfortunately, eliminating one competing hypothesis, like the S&W hypothesis of a complexity linking, only provides a moderate additional support to the AGW theory. This is because the “solar theory” is not a theory at all, but a fragmented set of mutually inconsistent hypotheses. If the S&W hypothesis (actually they have several) is falsified, there is a bunch of them left to keep the idea of the sun as the cause of recent decades’ global warming alive.
April 12 (13.27.24) James F. Evans write:
“Sometimes there is a thin line between proper scientific falsification and personal agendas.”
Our paper should be judged based on its scientific content, not on speculations about the authors’ personal agendas. It seems that you have nothing to say about the paper, only about why we wrote it. An it seems that your criticism is that we should have chosen another problem to work on. We ARE working on a lot of other problems, and most of them are not about climate change.
Here is how and why we wrote this paper. I am a plasma physiscist by training, I am approaching 60 years of age, and my son Martin (29) is a mathematician. During the last few years I have built a small cross-disciplinary group in complex systems science at the University of Tromso, north of the Arctic Circle in Norway. S&W work in a similar area, and this is how we learnt about their work. Since we work with similar methods we were able to read and understand these papers about complexity linking that to a great extent have been overlooked, presumably because it is hard for climatologists to penetrate the statistical physics jargon uused. We believe we found serious methodological weaknesses in the data analysis presented in those papers and decided to make our own analysis. We felt that our findings were interesting, and we decided to publish it.
So, what is wrong with that personal agenda?
April 12 (14.19.21) magicjava wrote:
“Let me see if I’m following along.
Scafetta and West analyzed data without de-trending it and it appeared to show that solar flares have a significant influence on global temperature. They appeared to be following the same random walk.
Rypdal and Rypdal analyzed the same data, but de-trended it and the correlation went away.
The concern is that by not de-trending the data, Scafetta and West unintentionally included factors other than solar flares in their analysis.
Correct?”
Let me try to give you in a nutshell what they did. They assumed the validity a model for the statistical fluctations of solar flares and global temperature called a Levy walk. This is different from a Levy flight (the diagram at the top of the page looks more like a Levy flight). The Levy walk model is fully described by two parameters, let us call them D and H (in the paper D is denoted by a greek delta and H by H_D). But there is really only one free parameter for the Levy walk, because for the Levy walk one has the relation
D=1/(3-2H). (1)
By fitting the solar flare data data to their model S&W found a value for D which is close to unity (D=0.94). The nature of the data the chose to analyze did not allow them to compute H directly. By applying a similar analysis to the global temperature (GTA) data without detrending they got D and H slightly different from each other, but both close to 1. That automatically makes Equation (1) above satisfied (D=H=1 satisfies the equation). One important property that distinguishes Levy walks on one hand and Levy flights and fBms on the other, is that Levy walks are not selfsimilar (D is different from H), but Levy flights and fBms are (H=D). Since they found H and D slightly different and approximately satisfying Equation (1) they concluded that the GTA is also a Levy walk, with roughly the same parameter D as the solar flare signal, and hence there should be a linkkage according to S&W.
By detrending the GTA we found D=H=0.65, and with a Gaussian distribution, which strongly indicates that it is an fBm. These values of D and H do NOT satisfy Equation (1), and hence it is NOT a Levy walk.
S&W have argued that one should not perform detrending. In an earlier message I have explained why one should.
oneuniverse (07:00:20):
If you’d bothered including the very next sentence, readers would immediately realise that the findings support Beer et al.’s 1988 results
Not to pollute the thread too much with OT stuff there is a couple of papers by Webber and Higbie that were discussed at length at the ISSI workshop:
http://www.leif.org/EOS/1003-4989.pdf
http://www.leif.org/EOS/1003-4989.pdf
The main conclusion: “These and other features suggest that galactic cosmic ray intensity changes which affect the production of 10Be in the Earth’s atmosphere are not the sole source of the 10Be concentration changes and confirm the importance of other effects, for example local and regional climatic effects, which could be of the same magnitude as the 10Be production changes.”
We should find another thread to discuss these if you wish to.
Kristoffer Rypdal (07:32:39) :
“By detrending the GTA we found D=H=0.65, and with a Gaussian distribution, which strongly indicates that it is an fBm.”
The problem should first be addressed from a mathematical point of view.
1) Detrending a Levy-Walk signal of its smooth component kills its Levy-Walk properties.
2) Levy-Walk signals may present a distribution of events that looks Gaussian.
Therefore R&R methodology is not appropriate for the task.
Leif Svalgaard (22:15:30) :
“The main argument of R&R, as far as I can tell, is that the PDF for the temperature does not have the sharp peaked structure of a Levy distribution.”
Leif Svalgaard (22:15:30) :
“The main argument of R&R, as far as I can tell, is that the PDF for the temperature does not have the sharp peaked structure of a Levy distribution.”
The PDF for the temperature does not need to have the sharp peaked structure of a Levy distribution at all! Levy Walk do not have such a peaked structure in their increments because they do not have long tails!
Kristoffer Rypdal (04:16:12)
Hello Dr. Rydpal, welcome and thank you for the explanations.
From the paper : “By subtracting a fitted fourth order polynomial from x(t), we obtain a detrended signal x_circumflex(t) which has Hurst exponent H ~ 0.65 in contrast to the H ~ 0.9 found in [1].”
Leaving aside the question of the appropriateness of detrending, may I ask what the reasoning is behind the choice of a fourth-order polynomial, over, say a 2nd order polynomial, or a 15th order polynomial ?
April 12 (19.52.00) Nicala Scafetta wrote:
“I suspect that it is another case similar to Benestad and Schmidt’s case. This authors have mistaken Levy-flights for Levy-walks. We are talking about Levy-walks not Levy-flights. Moreover, we have explicitly excluded in our papers a direct connection between the “increments” contrary to what Rypdal and Rypdal claim in their paper.
Levy-Walks are a property of the smooth component of a signal which emerge from a microscopic intermittency. Because of this, detrending the data in any way can destroy any memory associated with any statistics the data might have that develops in the time-frame dimension. Moreover, processed Levy-Walk signal may present increments that may indeed look like fractal Brownian motion.”
It is quite incredible that you can contend that we “mistake Levy flights for Levy walks”, when a major part of our paper is to point out the difference. I would appreciate if you could point out exactly which passages in our paper where we do these mistakes.
Fractional Brownian motions, Levy flights, and fractional Levy flights are well defined classes of stochastic processes and textbook stuff in the math literature. Levy walks are not that well defined. In your papers you actually use different definitions. In the analysis of solar flares you just look at event sequences characterized by discrete sequences of of times-the amplitude of events being irrelevant. These sequences are characterized by a distribution of waiting-times between events, and you characterize it as a Levy walk if this distribution has an algebraic tail with finite mean but infinite variance. Basically the waiting-times are Levy distributed, while for the Levy flight the increments are Levy distributed. You determine the exponent charcterizing the algebraic tail both by plotting the waiting-time distribution directly and by performing DEA analysis on the time-series obtained by considering the waiting time t_n versus the the event number n (the event number in the sequence functions as “time”). For numerically generated event sequences you also construct time series in “real” time by computing a frequency of events as a function of time. This time series can be analyzed both by DEA analysis and SDA analysis. You also present another version where a time series in real time is constructed by assuming that the walker moves with a given velocity in either the positive or the negative direction. At the times of the events the sign of this velocity is redefined by the flipping of a coin.
We have shown that the SFI time series is a Levy FLIGHT, at least on time scales longer than a few months. We can produce an event sequence from this signal by introducing a threshold on the signal amplitude, and recording the times when the signal amplitude rise above that threshold. If the threshold is relatively low, we get many waiting times that are shorter than a month, and the waiting-time distribution will be algebraic for these short scales, but it will become exponential for waiting times larger than a month. Exponentially distributed waiting times implies Poisson statistics, i.e. the event probability is independent on the time since the last event (no memory). These results are completely consistent with the solar flare statistics performed in your papers, you just avoid mentioning the important fact that there is no memory in the solar flare statistics on time scales longer than a few months.
When we turn to the GTA signal you don’t have an explicit event statistics and it is not so meaningful to define events from using thresholds. So I assume that you envisage that this time series is modeled by one of the “real time” time series mentioned above, and you perform DEA and SDA analysis on this time series.
Please, Nicola, point out where I have misunderstood.
The crucial question is then if the slow, multidecadal trend in the GTA signal should be considered as an inherent part of the Levy walk, or whether it should be subtracted to discern the contended Levy-walk statistics. We believe it can be easily demonstrated that this trend must be removed. This will be the subject of my next message. But first I need some sleep (the time is 23.17 in Norway).
I apologise for my mistaken spelling of your name Dr. Rypdal, please forgive me.
Kristoffer Rypdal (14:18:34) :
“Exponentially distributed waiting times implies Poisson statistics, i.e. the event probability is independent on the time since the last event (no memory).”
This isn’t what was observed in Leif’s colleagues’ ‘Science Nugget’.
http://arxiv.org/PS_cache/arxiv/pdf/1002/1002.4869v1.pdf
From this I deduce that the temporal duration between flares determines the amplitude of the flare (e.g.. A longer time scale between flares evidences a greater flare eruption amplitude, whereas a shorter time scale between flares evidences a lesser flare eruption amplitude). Thus, there is something that remembers the timing of the last flare eruption and this is presented in the amplitude of the following flare eruption. This is a ‘memory’!
However, climate also provides a second ‘memory register’. Though this lags events simply because it needs to be ‘taught’ how to respond. Ozone in the stratosphere is altered (generated) by the insolation of EM radiation at wavelengths of ultraviolet and shorter into the ‘ozonosphere’. Although the ‘ozonosphere’ has a relatively quick ‘turnover’ of specific O3 gas molecules (ozone is relatively short lived), it is likely to permit short bursts of energy through while it builds ozone, but as energy bursts decrease the ‘ozonosphere’ provides an excellent barrier to ionising wavelengths (it’s a ‘learning and forgetting curve’).
As I said, a second ‘memory’ to the chain of events for GTA. How is it that you see none when I see, at least, one at each end of this interaction?
While I’ll admit that I’m out of my depth in this thread and I’m here for ‘learning purposes’, but this is what I see here.
BTW, it’s now 01:45+ here in the UK at this time of posting. Good night and god bless (just a local phrase). 🙂
Best regards, suricat.
suricat (17:51:39) :
From this I deduce that the temporal duration between flares determines the amplitude of the flare (e.g.. A longer time scale between flares evidences a greater flare eruption amplitude, whereas a shorter time scale between flares evidences a lesser flare eruption amplitude). Thus, there is something that remembers the timing of the last flare eruption and this is presented in the amplitude of the following flare eruption. This is a ‘memory’!
Not really, as great flares simply are much rarer than small flares. The might be a slight ‘memory’ in the small flares, because a given active region often produces many small flares. This is not a real memory, as each flare is basically independent of the previous one: the magnetic field gets wound up by plasma motions and gains energy by this. If the energy exceeds a threshold [set by local conditions, e.g. the shape of the field], the flare blows. It is like stretching an elastic until it breaks, then stretching another elastic until it breaks, etc. The point where the second elastic breaks does not depend on where the first elastic broke.