Guest Post by Willis Eschenbach
Over at Pierre Gosselin’s site, NoTricksZone, he’s trumpeting the fact that there are a bunch of new papers showing a solar effect on the climate. The headline is Already 23 Papers Supporting Sun As Major Climate Factor In 2015 …Burgeoning Evidence No Longer Dismissible!, complete with exclamation mark … sigh. Another person who thinks that because a paper is published in a scientific journal it’s not “dismissible” … skeptics of all people should know better than that. In any case, I figured I should at least take a look, and so as not to pick favorites, I grabbed the first paper in Pierre’s list.
Turns out that the very first paper was one of two discussed back in January here on WUWT. I didn’t see the WUWT post at the time, so it’s now my sad duty to pick up my shovel, put on my hip-boots, and wade into the mire.
Both papers are by David Douglas and Robert Knox. David has published occasionally here on WUWT. So I grabbed that first paper, yclept “The Sun is the climate pacemaker I. Equatorial Pacific Ocean temperatures”, available here.
I must confess that their Abstract left me scratching my head … it says:
Equatorial Pacific Ocean temperature time series data contain segments showing both a phase-locked annual signal and a phase-locked signal of period two years or three years, both locked to the annual solar cycle. Three such segments are observed between 1990 and 2014. It is asserted that these are caused by a solar forcing at a frequency of 1.0 cycle/yr. These periodic features are also found in global climate data (following paper). The analysis makes use of a twelve-month filter that cleanly separates seasonal effects from data. This is found to be significant for understanding the El Niño/La Niña phenomenon.
They claim that there are climate phenomena that are “locked to the annual solar cycle” … say what? Almost every climate phenomenon I know of is locked to the annual solar cycle with some variable amount of delay. How is that possibly news? I didn’t get that when I read it, and re-reading the paper hasn’t helped much.
Upon first reading, I thought that the secret might lie in the fact that they say they have a filter that “cleanly separates seasonal effects from data”. So maybe they’re not talking about just being phase-locked to the plain old solar cycle … because they’re looking at the data after the seasonal effects have been removed in some special way.
Now I have some interest in filters, so I looked to see what they were talking about. The normal way to remove the “seasonality”, the month-by-month changes in temperature, is to take monthly averages and subtract them from the data. These monthly averages for January to December are usually called the “climatology” for the region. However, the authors don’t like that process for some reason. They describe their own procedure as follows:
2.2. Methods: precise separation of high- and low-frequency effects
Studies of many geophysical phenomena start with a parent signal G0, such as a temperature or wind speed record, containing a component of interest mixed in with a seasonal component at frequencies of 1.0 cycle/yr and its harmonics. The component of interest might show ENSO effects with multi-year periodicity. An important task is to separate the seasonal component from G0 to obtain the one of interest. A moving average is one of the methods used to make this separation. Such a filter of length one year, which we denote by an operator F, is the most precise for seasonal components, as Douglass [5] has shown in a study of SST3.4 (the parent signal).
OK, they’re using a 12-month moving average filter, often called a “boxcar” filter, to remove the climatology. Not a choice I’d make, because it messes with the data at other cycle lengths. Let me demonstrate this problem in two ways. First, here is the underlying data (gray line in background), along with same data with the climatology (monthly averages) removed (blue line with dots), and finally in pale red, the results using their 12-month moving average “boxcar” filter.
Figure 1. The monthly sea surface temperatures for the “Nino 3.4″ region, which extends 5° north and south of the equator from 120°W to 170°W in the tropical Pacific Ocean. The gray line in the background is the original data. The blue line with dots shows the normal method of removing the seasonal variations, by subtracting the monthly averages (the “climatology”) from the original data. The pale red line shows the result of applying their 12-month moving average “boxcar” filter to the original data.
What is clear from Figure 1 is that the 12-month boxcar filter (red line) is removing much more of the variation in the data than is the normal method of subtracting the climatology (blue line). The red line has smoothed away much of the short-term fluctuation in the original data. And while this result of theirs is of interest, it is not accurate to claim as they do that:
Such a filter of length one year, which we denote by an operator F, is the most precise for seasonal components …
I said I would show the boxcar filter problem in two ways. Here’s the other way to illustrate the difference between climatology and boxcar filter, using a periodogram. A periodogram shows the strengths of the various cycles that make up the signal. Here are the periodograms of the three different datasets used in Figure 1:
Figure 2. Periodogram showing the cycle strengths of the original SST3.4 data (gray, only visible at cycles of one year or less), the original data with climatology removed (blue), and the 12-month moving average “boxcar” filter of the original data favored by the authors (red).
As one might expect, the original data (gray line) shows a strong cycle at both one year and six months (0.5 years). The precision of the climatology method (blue line) is shown by the fact that it only affects the data at frequencies of one year or less. Above that, you can’t even see the original data (gray line) because other than a slight difference at 13 months, the original gray line is hidden by the blue line, meaning that subtracting the climatology has not affected the cycles of other lengths at all.
Now contrast that with the effect of their filter (red line). In addition to doing what the authors desired, that is to say removing the one year and six month peaks, it has an unwanted side-effect. It has greatly reduced the strength of the cycles between one year and six years or so as well. This is the same thing we saw in Figure 1, where their boxcar filter (red line) was smooth and did NOT show the short-term cycles.
Setting that difficulty aside, let me move on to what they do with their boxcar filtered data. Their main scientific claim is that even though they’ve removed the annual variations with their boxcar filter, there are still some time periods that show evidence of either a two-year or a three-year cycle. They call these intervals “phase-locked” with the sun. Here’s their graph detailing those periods:
Figure 3. This shows Figure 3a from their paper. The thick black lines at the bottom indicate what they call “previously reported climate shifts”, although they give no citation for where they were “previously reported”. My guess is that these “climate shifts” were “previously reported” by the authors themselves, but then I’m a skeptical fellow. ORIGINAL CAPTION: Fig. 3. a. Low frequency index aSST3.4 (red) and NOAA anomaly index Nino3.4 generated by the climatology method (blue).
Before discussing this Figure, a momentary digression. Here is a giant red flag from their paper:
2.1. Data
This study considers only data from the period January 1990 through December 2013.
They are not using all of the data. The dataset is already short, only from January of 1982 through December 2013 at the time of their writing the study, or a total of 32 years of observations. Despite that, they’ve thrown away no less than eight years of the data, a full quarter of the available information … why? Unfortunately, the only discussion of that question I could find in their paper is the short sentence I quoted above.
Now, when someone does that, my urban legend alarm goes off. It generally means that the data are being stubborn and uncooperative … but I digress, back to Figure 3.
Looking at that Figure, I gotta say … whaaaa? This is their finding that justified publication in a scientific journal? This is the sum total of the first paper of the widely-hyped TWENTY-THREE NEW PAPERS ON SOLAR blah blah blah? This is it?
If you don’t see why I am so incredulous, there are several reasons. The first reason is the length of the time periods. Let’s take a look at interval #11 at the right of their Figure 3. It covers 5 3/4 years … and they are claiming a 3 year “phase-locked” cycle exists there? That’s not even two full cycle lengths! In climate science you need to have four cycle lengths to say whether a cycle is real and persistent or not … and often even that is sometimes not enough. And here, they are declaring a cycle is alive and well on the basis of not even two full cycles of data. This is meaningless.
And the same is true for time intervals #9 and #10 above. Neither of them are even three cycles in length. Declaring the existence of a “phase-locked” interval on that basis is foundation-free.
Second, what did they expect? If you have a complex signal like the SST3.4, it has cycles of a variety of lengths going on in it. Take a look at the periodogram in Figure 2 above. The original data has all kinds of cycle lengths which have some strength even after removing the annual signal by either method. So OF COURSE there are sections which have a stronger two-year or a three-year cycle in them, just as there are sections with stronger or weaker 2.5 year cycles in them.
Third, they make much of the fact that the cycles they’ve found are exact-year periods. Yes, those exact-year cycles are there, but per the periodogram, they’d do better by looking for cycles at 2.5 years, 3.75 years, and 5.5 years …
Here’s the problem. No matter what length cycle you look for … you’ll find it. So their claim that the periods are locked or related to the sun are nonsense—they’ve only looked at sun-related (exact-year) timespans, and surprise, that is what they have found.
Having digested all of that, I had to ask … so what?
Seriously, so what? What does one get by rooting through a pile of cycles and choosing some to focus on?
Well, here’s their answer to the question about why all this matters:
This study confirms the results of [1] that some of the largest maxima/minima in the oscillations of the phase-locked state correspond to well-known El Niños/La Niñas. For example, the sequence 1996 La Niña – 1997/98 El Niño – 1999 La Niña corresponds to a minimum–maximum–minimum portion of phase-locked segment #9.
Before I get into El Ninos, this quote brings up an issue that has bugged me throughout, similar to the issue of their omitting eight years of data … where are the “phase-locked segments” numbers #1 through #8? How come they didn’t show them or say one word about them? And since there is eight years of missing data, it doesn’t seem possible that the “phase-locked segments” #1 through #8 could be there. In any case, one rule that has rarely failed me, in climate science as in life, is that when a man hides something … it means he’s got something to hide. But again I digress … back to the El Ninos.
Their claim is that some of the largest maxima and minima in the El Nino 3.4 index correspond with El Ninos and La Ninas … again, I was dumbfounded. Large maxima in the El Nino 3.4 region correspond with large El Ninos? Who would have guessed? Why do the authors imagine that it’s called the “Nino3.4” region?
In any case, given their claims above about El Ninos, it appears that the scientific value of the 2-year and 3-year so-called “phase-locked sections” is to understand and thus better predict El Ninos. And to be sure, new theories can indeed have value if they can make testable predictions, regardless of how outré their claims or explanations might seem. Sooo … here is their daring prediction based on their work:
The climate system is presently (June 2014) in a phase-locked state of periodicity 3 years. This state, which began in 2008, contains a maximum (El Niño) at about 2010 followed by a minimum (La Niña) followed by a maximum (weak El Niño at about 2013). If the climate system remains in this phase-locked state, the next maximum will not occur until about 2016 – i.e., no El Niño before that date. On the other hand, if a maximum occurs before then, it will signal the end of the phase-locked segment (and therefore a climate shift).
I gotta admit, I lost the plot entirely when I read that. If the climate system stays “phase-locked” it means an El Nino at the next maximum, unless no El Nino occurs at the next maximum, in which case it means a climate shift.
Given that the data they are using is SSTs of the the Nino 3.4 region, and given that El Ninos are defined inter alia by maxima in the sea surface temperature anomalies in the El Nino regions … I don’t even know what that prediction means. The only thing I can compare it to is Will Rogers’ unbeatable formula for making money in the stock market:
Buy a stock, and when it goes up, sell it. And if it doesn’t go up, don’t buy it.
…
Onwards to their conclusions, I can’t resist one more quote:
6. Conclusions and summary
Phase-locked sequences are found in Pacific Ocean SST3.4 temperature data during the periods 1991–1999, 2002–2008 and in 2009–2013. These three sequences apparently being separated by climate shifts. It is asserted that the associated climate system is driven by a forcing of solar origin that has two manifestations: (1) A direct phase-locked response to what is identified as a solar forcing at a frequency of 1.0 cycle/yr for the whole time series;
I couldn’t make it to the second “manifestation”, I was laughing so hard. It is boldly “asserted” that the temperature of the Pacific Ocean is “phase-locked” to “what is identified” as “a forcing of solar origin”??? You mean that the ocean temperature follows the sun? Who would have guessed? Who was the genius that first identified that it was “a forcing of solar origin”? That definitely proves that the sun has an effect on the climate, all right, no gainsaying that …
All kidding aside, let me put something on the table. First, it’s obvious that the sun affects the climate. Without the sun, we’d be pretty cold. And yes, according to this paper the temperature has what is usually called an “annual cycle” but which they refer to as a “phase-locked response to what is identified as a solar forcing at a frequency of 1.0 cycle/yr” … However, related to cycles longer than one year, things get murky pretty fast. In particular, consider the long-time hunt for some sign of the ~11-year solar sunspot-related cycle on the climate.
We humanoids have been looking for a definite clear effect on the climate that could be attributed to the solar variations associated with the sunspot cycle ever since William Herschel made his failed prediction (see below) about sunspots and wheat prices a couple of centuries ago. If such a clear definite effect had ever been demonstrated, we wouldn’t be still having this discussion. After hundreds and hundreds of people starting with Herschel and up to and including myself and others have looked over a total period of two centuries for evidence of such an effect, one thing is clear:
If something associated with the ~11-year sunspot cycle is having an effect on the climate, it is a very small effect, otherwise it would have been both identified and verified beyond question years ago.
At this point, the hunt for such evidence has become so obsessive that I was seriously presented with a paper that the commenter assured me clearly demonstrated that something associated with the ~11 year sunspot cycles was indeed having a measurable effect on the climate. It turned out the that evidence was in the form of tree ring records … tree ring records from one single core from one single tree in Chile.
One Chilean tree! That’s how desperate some folks are to have their ideas validated … and how desperate the scientific journals are for things to publish.
Now, given the number of One Chilean Tree papers published each year, including this paper discussed above, there’s no way that I could possibly deconstruct them all. First off I have to read and understand their paper. Then I have to go get the data they used and replicate their study, as I did above for this study. I have to do my own analyses until I’m clear where they’ve gone off the rails. Then I have to produce the graphics, which better be error-free, and write the paper, which hopefully is error-free or I will be properly and quickly (and fortunately) informed of my mistake(s).
Finally, I have to upload the paper to the web, upload all the graphics, connect up all the links, tag it and categorize it, spell-check it, and proof-read the preview to make sure it’s all correct. Oh, and pick the featured image, can’t forget that. From your side it just magically appears on your screen … on my side, each one is a pile of work.
So I’m declaring right now, I’m not touching the other 22 papers listed by Pierre. At this point, the onus is on you. I’m just one guy, no graduate students or associates, I can’t stem the flood of Chilean trees. So … if you think that something associated with the sunspot cycle (TSI, EUV, solar wind, GCRs, heliomagnetic field, pick your poison) is having an effect down here at the surface of the earth where we live, and you think you have the scientific paper that conclusively demonstrates it, then you are welcome to send me TWO LINKS:
• A link to a non-paywalled version of the paper. I’m not paying $37 to read about another Chilean tree.
• A link to the exact dataset(s) used by the authors in their study.
Don’t bother me with data dumps of five or twenty-three papers, not interested. I want the one paper that YOU think is the best, second place doesn’t interest me. I won’t guarantee to write about whatever paper it is, but I will write about it if the data and the analysis stands up. Remember that one link is not sufficient. I need a link to both the non-paywalled paper and to the data they used. Please, no papers about solar effects on the thermosphere or the Van Allen belts, read my request again.
Best to all,
w.
PS—In these parlous times, if you disagree with someone (unlikely, I know, but it happens), please quote the EXACT WORDS YOU DISAGREE WITH. This allows all of us to know both who you are addressing, and what specifically what you are objecting to.
HERSCHEL: Before you get all steamed up and start yelling at me about how you know for a fact that the astronomer William Herschel proved that wheat prices varied with the sunspots, read On the insignificance of Herschel’s sunspot correlation, published in Geophysical Research Letters. I’d written a post on the subject a couple years ago that came to the same conclusion, but I never published it because I came across that link, and the author did it so much better. If you have specific problems with that paper, feel free to list them. While you are at it, you might profitably contemplate the concept of “scientific urban legends” …
FURTHER READING: If you have not done so, you might enjoy reading my previous posts on the sunspot-cycle question …
Riding A Mathemagical Solarcycle 2014-01-22
Among the papers in the Copernicus Special Issue of Pattern Recognition in Physics we find a paper from R. J. Salvador in which he says he has developed A mathematical model of the sunspot cycle for the past 1000 yr. Setting aside the difficulties of verification of sunspot numbers for…
Congenital Cyclomania Redux 2013-07-23
Well, I wasn’t going to mention this paper, but it seems to be getting some play in the blogosphere. Our friend Nicola Scafetta is back again, this time with a paper called “Solar and planetary oscillation control on climate change: hind-cast, forecast and a comparison with the CMIP5 GCMs”. He’s…
Cycles Without The Mania 2013-07-29
Are there cycles in the sun and its associated electromagnetic phenomena? Assuredly. What are the lengths of the cycles? Well, there’s the question. In the process of writing my recent post about cyclomania, I came across a very interesting paper entitled “Correlation Between the Sunspot Number, the Total Solar Irradiance,…
Sunspots and Sea Level 2014-01-21
I came across a curious graph and claim today in a peer-reviewed scientific paper. Here’s the graph relating sunspots and the change in sea level: And here is the claim about the graph: Sea level change and solar activity A stronger effect related to solar cycles is seen in Fig.…
Sunny Spots Along the Parana River 2014-01-25
In a comment on a recent post, I was pointed to a study making the following surprising claim: Here, we analyze the stream flow of one of the largest rivers in the world, the Parana ́ in southeastern South America. For the last century, we find a strong correlation with…
Usoskin Et Al. Discover A New Class of Sunspots 2014-02-22
There’s a new post up by Usoskin et al. entitled “Evidence for distinct modes of solar activity”. To their credit, they’ve archived their data, it’s available here. Figure 1 shows their reconstructed decadal averages of sunspot numbers for the last three thousand years, from their paper: Figure 1. The results…
Solar Periodicity 2014-04-10
I was pointed to a 2010 post by Dr. Roy Spencer over at his always interesting blog. In it, he says that he can show a relationship between total solar irradiance (TSI) and the HadCRUT3 global surface temperature anomalies. TSI is the strength of the sun’s energy at a specified distance…
The Tip of the Gleissberg 2014-05-17
A look at Gleissberg’s famous solar cycle reveals that it is constructed from some dubious signal analysis methods. This purported 80-year “Gleissberg cycle” in the sunspot numbers has excited much interest since Gleissberg’s original work. However, the claimed length of the cycle has varied widely.
Cosmic Rays, Sunspots, and Beryllium 2014-04-13
In investigations of the past history of cosmic rays, the deposition rates (flux rates) of the beryllium isotope 10Be are often used as a proxy for the amount of cosmic rays. This is because 10Be is produced, inter alia, by cosmic rays in the atmosphere. Being a congenitally inquisitive type…
The Effect of Gleissberg’s “Secular Smoothing” 2014-05-19
ABSTRACT: Slow Fourier Transform (SFT) periodograms reveal the strength of the cycles in the full sunspot dataset (n=314), in the sunspot cycle maxima data alone (n=28), and the sunspot cycle maxima after they have been “secularly smoothed” using the method of Gleissberg (n = 24). In all three datasets, there…
It’s The Evidence, Stupid! 2014-05-24
I hear a lot of folks give the following explanation for the vagaries of the climate, viz: It’s the sun, stupid. And in fact, when I first started looking at the climate I thought the very same thing. How could it not be the sun, I reasoned, since obviously that’s…
Sunspots and Sea Surface Temperature 2014-06-06
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…
Maunder and Dalton Sunspot Minima 2014-06-23
In a recent interchange over at Joanne Nova’s always interesting blog, I’d said that the slow changes in the sun have little effect on temperature. Someone asked me, well, what about the cold temperatures during the Maunder and Dalton sunspot minima? And I thought … hey, what about them? I…
Splicing Clouds 2014-11-01
So once again, I have donned my Don Quijote armor and continued my quest for a ~11-year sunspot-related solar signal in some surface weather dataset. My plan for the quest has been simple. It is based on the fact that all of the phenomena commonly credited with affecting the temperature,…
Volcanoes and Sunspots 2015-02-09
I keep reading how sunspots are supposed to affect volcanoes. In the comments to my last post, Tides, Earthquakes, and Volcanoes, someone approvingly quoted a volcano researcher who had looked at eleven eruptions of a particular type and stated: …. Nine of the 11 events occurred during the solar inactive phase…
Early Sunspots and Volcanoes 2015-02-10
Well, as often happens I started out in one direction and then I got sidetractored … I wanted to respond to Michele Casati’s claim in the comments of my last post. His claim was that if we include the Maunder Minimum in the 1600’s, it’s clear that volcanoes with a…
Sunspots and Norwegian Child Mortality 2015-03-07
In January there was a study published by The Royal Society entitled “Solar activity at birth predicted infant survival and women’s fertility in historical Norway”, available here. It claimed that in Norway in the 1700s and 1800s the solar activity at birth affected a child’s survival chances. As you might imagine, this…
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
Willis,
Thank you for all your efforts at going through these much heralded papers. As just a lowly B.S. degreed engineer, you do a great job of making sense of all the technical obfuscation that appears to be the norm for climate science nowadays. The graphs you painstakingly produce really illustrate where the authors miss the boat.
Response to Willis Eschenbach’s comments on the Douglass/Knox paper “The Sun … I”
Almost all of the comments by Eschenbach in regard to the Douglass/Knox paper “The Sun is the Climate Pacemaker I …” are wrong. We (Douglass/Knox) will reply to a few of them.
————————————–
“Almost every thing I know is locked to the annual Solar cycle…”
Reply:
Most know about the seasonal effect (winter vs summer) at high latitudes caused by changes in solar irradiance whose average is 342W/m2. However, there is another effect at 1 cycle/year caused by the eccentricity of the earth’s orbit. The amplitude of this solar irradiance is 11.3 W/m2 and peaks in early January. This solar forcing is the “Pacemaker” as described in the 2nd paragraph of section 2.2 of our paper.
——————————————
“… they give no citation for where were ‘previously reported’ . My guess is that these ‘climate shifts’ were ‘previously reported’ by the authors themselves…”
Reply:
No, they are from many sources and are described in a paper by Douglass (reference 1). In reference 18 in this paper a list of climate shifts is given from various studies. [Note. As every scientist knows, references are given so that the reader may know where an assertion of the author may be checked. With the help of Google any reference can be downloaded in less than a half hour. For the Douglass/Knox papers there is link just below the references. One click and a list appears. A second click gives you the paper in question. All in less than a minute.]
————————————————-
“They are not using all of the data. … It generally means the data are being stubborn and uncooperative.”
Reply:
The data are not being stubborn and uncooperative – just the opposite. The missing data from 1982 to 2000 is the same as seen in figure 2 of reference 1. To include this data would bring in phase-locked segment #8 which would require additional redundant discussion. That is why we began at 1990.
———————————
‘”precision of the climatology method”
Reply:
Section 4 is entirely about how the “climatology” method is flawed.
—————————————-
Where are “…phase-locked segments #1 through #8” ?
Reply:
Two clicks away in reference 1, as mentioned specifically in the Introduction. All of them are clearly identified in figure 2. Also in this figure the “climate shifts” are shown as blue arrows.
———————————————
“ … cycles at 2.5 years, 3.75 years, 5.5 years…”
Answer: The peaks for periods greater than 1 year in Eschenbach’s periodgram are meaningless. They are some kind of average of segments of 2-year, 3-year and segments containing neither.
———————————————
David Douglass
Robert Knox
“One click and a list appears. A second click gives you the paper in question. All in less than a minute.”
Willis’ unwillingness to obtain papers and data himself reminds me Maynard G. Krebs’ aversion to work. He also doesn’t observe the usual form for writing scientific papers.
Thanks for clearing up the supposed mystery of the allegedly “hidden” data. Much as so many here expected, no offense committed.
Because…male chauvinism. Right?
No. In that case, it’s laziness and contempt for the scientific method. It’s all about the megalomania and desperate attempt to achieve something at the end of a wasted life. IMO.
IIRC, Willis has a healthy, happy daughter, in which case, not totally wasted.
Oh. And your acumen in the scientific method is superior? Please demonstrate. Spit out an ANOVA with just a calculator. Hell. Let’s up the stakes. Make it an ANCOVA. Here’s a link for what I consider to be akin to ANOVA’s For Dummies.
http://www.statsmakemecry.com/smmctheblog/stats-soup-anova-ancova-manova-mancova
Yes, Pamela, not only my grasp on the scientific method, but elementary math and statistics. Willis, the rank newbie (not amateur, since he would have to love statistical analysis for that term to apply), has been repeatedly shown up for the poseur he is here by real experts in statistical analysis.
This AW’s blog, so of course he’s free to publish whatever he wants, but Willis’ “analyses” are so laughably wide of the mark that I sometimes wonder if he isn’t a Mannian plant. I refer you, just for a single ludicrously erroneous post, to the evisceration the psych major Willis suffered at the expert hands of 1sky1 and Dinostratus (if I remember their handles correctly) on his risible NAO “analysis”.
To say nothing of his pretending to have discovered tropical oceanic cloud formation as a weather phenomenon, which Gore-like claim was deflated by a real expert, Dr. Spencer.
IMO, AW should rid himself of this obnoxious poseur.
Lady GG,
Sorry, and I may well and perhaps not without reason, be censored for this comment, but in my considered opinion, Willis is a statistical stupe, a mathematical moron and arithmetic a$s.
As has been amply demonstrated not only by Dr. Douglass here, but in every single one of Willis’ lame attempts to dispute the glaringly obvious fact of solar influence on climate.
If moderators want to silence me, I’ll understand.
Again, childish, interruptive comments instead of showing your work. Put up your research. And if you want to go head to head with it, I’ll put up mine. Granted, it’s not in climate. It’s on the Auditory Brainstem Response to high frequency tone pips. My previous career. While I am a one-hit wonder in published research, that one was a hit out of the park. Since then, instead of studying brains, I grow them. Even the ones that don’t work so well.
Insolation, irradiance. Same thing. Right David? No?
David, I have never and would never cite a paper that cites references that are germane to my thesis, in order to cite references germane to my thesis. I would cite those papers directly with comments.
People can discuss, dispute, accept or reject the science that indicates the influence of solar activity dominates when it comes to climate change. But what is undeniable is that over the last 2000 years, there is good correlation between solar activity and climate change while there is no correlation between carbon dioxide and those major climate changes over the last 2000 years.
Over the last 2000 years, a period in which atmospheric carbon dioxide was relatively stable, there has been the Roman Warm Period, the Dark Ages Cold Period, the Medieval Warm Period, the Little Ice Age, and the recent warming trend. Very cyclical… just as solar activity is very cyclical.
So … I am inclined to believe that the key influence on earth’s planet is very likely the sun and that the IPCC’s mantra is utter rubbish as demonstrated by Dr Giaever’s presentation in the following video:
A brave man and a great scientist, but I wish he hadn’t said “degrees K” instead of just “K”.
Also, chimps barter bush baby meat in exchange for sex. Humans aren’t the only bartering species, contrary to Matt Ridley’s view. Otherwise, I’m totally down with Matt.
PS: However, not much in the video about the effect of the sun on climate, which is the topic under discussion in these comments. It is however devastating to the repeatedly falsified hypothesis of CO2 as the control knob on climate.
William Astley
September 23, 2015 at 1:06 am
—————————————–
Interesting William. Changes in the Earth’s magnetic field in particular.
Aren’t we due for a “Solar Cycle Up Date” here at the WUWT?
As Mosh observed above, not only does Lady Gaiagaia struggle with the word “two”, I feel that the meaning of the word “data” is completely foreign to her. Hint: It’s plural.
This is quite depressing for me. As I observed in an email to Willis:
“It is unfortunate that so many of my ideological allies are such dimwits. It makes us all look bad in very serious ways.”
~ ctm
I’m quite sure that my understanding of Latin is at least as profound as yours.
Why is it that on this blog the obviously math challenged megalomanic Willis gets to chose which data sets to which he will reply, while ignoring others?
If you are typical of moderators here, then anyone seriously interested in science, to say nothing of math and statistics, should avoid this site like the plague.
Thanks for making clear to me that avoiding science is the purpose of this blog.
Bye bye and ta ta.
But please be aware that real scientists and genuine statisticians laugh at you in general and the presumptuous ignoramus Willis in particular.
Charles, data are mostly singular: Google has 159M hits for “data is”, only 119M for “data are”.
David Douglass September 23, 2015 at 3:34 pm
David, thank you for your genteel responses, perhaps better than I deserved. My comments below.
I fear I’m not following this. The “2nd paragraph of section 2.2” says in toto:
This says nothing about a “Pacemaker”. So I did a search for “pacemaker”, and the only mention of a “pacemaker” is in the title of your study. Nor is eccentricity mentioned anywhere in your paper, not once.
I think you must mean the second paragraph of section 3.2, although that doesn’t really help much either. It simply claims that the variation in SST3.4 temperatures is due to TSI, which seems reasonable. It says:
Unfortunately, this isn’t true. The actual situation is much more complex than the one you describe. You are correct that for the entire globe the total solar insolation (TSI) peaks in January and June. Here are the observations, the seasonal variations, and the (very small) residuals from the CERES data. The January/June peaks are quite visible.
Figure S1. Incoming top-of-atmosphere total solar irradiation (TSI) for the globe. Top panel shows observations. Middle panel shows the repeating identical seasonal component. The bottom panel shows the residuals, the deviation of the observations from the seasonal expectation.
The top panel shows the ± 11 W/m2 swing that you discuss in Paragraph 2 of 2.2 above. You can see the variation in TSI linked to the sunspots in the bottom panel. So far, so good.
However, around the Equator, things get strange. Half the year the sun is in the north, half the year it’s in the south. In addition to the annual cycle you discuss, this imposes a half-year cycle as well. This changes the situation considerably. Overall, the variation in TSI in the equatorial Nino3.4 region looks like this:
Figure S2. Incoming top-of-atmosphere total solar irradiation (TSI), Nino 3.4 region. Top panel shows observations. Middle panel shows the repeating identical seasonal component. The bottom panel shows the residuals, the deviation of the observations from the seasonal expectation.
As you can see, the TSI over the Nino3.4 region is absolutely nothing like you’ve described above. First, the six-month signal is quite large. How large? I’ll get to that.
More to the point, the peak insolation in the Nino3.4 region is NOT in January, as you state in your paper. It is in March. But wait, it gets worse. The variations in TSI are wildly asymmetrical. The peak insolation is in March … but the minimum insolation is only three months later, in June.
Next, as to the question of which cycle is larger—the six-month cycle due to the sun moving north/south, or the one year cycle as the earth moves closer and further from the sun? TSI falls off as the cosine of the zenith angle. The sun moves 23.45° North and South over the year. Measured at the Equator at the time when the sun is furthest from the Equator, TSI is 416 W/m2 * cos(23.45) = 381 W/m2. This is a variation of ± 17 W/m2, compared to your accurate quotation of ± 11 W/m2 for the variation due to eccentricity. This difference in strength is also visible in the periodogram of the Nino3.4 TSI:
Figure S3. Periodogram, top-of-atmosphere total solar irradiation (TSI), Nino 3.4 region
Here you can see the two cycles, six months and one year. And as you can see, the 6-month cycle is the stronger.
To sum up: The “Pacemaker” you have described is nothing like the reality. Rather than having a peak in January and a minimum in July as you claim, it has a peak in March and a minimum in June. And rather than the strongest cycle being the annual eccentricity cycle as you claim, the strongest (and fastest changing) cycle is the six-months-north, six-months-south familiar annual cycle.
Now, here’s where it gets very interesting. As we’re all aware, the ocean is not at the top of the atmosphere. So I decided to look at what is actually happening down at the surface. After all, in general that’s what the sea surface temperature swings are most related to—not TSI, but how much sunlight they actually receive at the surface. Fortunately, that information is available in the CERES data. As usual, I was surprised by my findings. First, here’s an overview of the full period of record:
Figure S4. Absorbed surface solar radiation, Nino 3.4 region. Top panel shows observations. Middle panel shows the repeating identical seasonal component. The bottom panel shows the residuals, the deviation of the observations from the seasonal expectation.
Now, when I looked at Figure S4, my jaw dropped. I looked at the bottom graph, and I thought, oh my goodness, that looks very much like the inverse of the SST3.4, the very data we’ve been discussing. SO … I ran to calculate the CCF, the cross-correlation function between the absorbed solar anomaly and the sea surface temperature anomaly (climatology removed in both cases) … here’s that graph:
Figure S5. Cross-correlation of the absorbed surface solar radiation anomaly and the temperature anomaly, Nino 3.4 region. Climatology removed in both cases. The lag shows the amount that the reduction in solar lags the temperature change.
And that, to me is a most lovely graph. I was correct in that the absorbed solar anomaly is indeed the inverse of the temperature anomaly. This is among the strongest observational evidence for my thermostat hypothesis that I have ever found. What a surprise to find it when I was least expecting it!
The reason I was excited is that this shows is that as the sea surface temperature (SST) anomaly warms in the Nino3.4 region, the downwelling solar decreases. Decreases! The surface is getting warmer, and in response there is less solar radiation at the surface.
I say that this is because of increased clouds from the warmer surface reducing the incoming radiation, just as is predicted by my thermostat hypothesis.
So David, I find I have to thank you. This discussion with you has led me to this very strong evidence, which otherwise I might never have found … life is a strange thing.
To close out this part, let me show a detailed view of Figure S4, the variations in surface absorbed radiation.
Figure S7. Closeup of Figure S4Absorbed surface solar radiation, Nino 3.4 region. Top panel shows observations. Middle panel shows the repeating identical seasonal component. The bottom panel shows the residuals, the deviation of the observations from the seasonal expectation.
As you can see, once again there has been a transformation of the previous cycle. Whereas the TSI peaked in March and was lowest in June, at the surface the relative size of the two peaks is swapped. Now the peak in surface insolation is in October, and the minimum is still in June … go figure.
In any case, that’s the reply to the first of your comments, and it certainly didn’t end up going the direction I thought it would.
In that regard, let me leave you with a final pair of graphs. These show the decomposition of the SST3.4 temperature itself, the subject of your paper.
and here is a closeup of the recent period:
It was the peaks in 2003, 2007, and the big peak in 2010 that made me think that the bottom panel of Figure S4, the absorbed solar, might be the inverse of the SST. It has valleys at the same points.
As you can see, the SST peaks in May as you’ve said … but the actual situation is more complex than that. The peak is in May, but the minimum is not until eight months later, in January.
Finally, where in all of that confusion of a combination of a larger 6-month cycle and a smaller annual cycle are we supposed to find your “Pacemaker”? Things are nowhere near that simple.
Thank you again for your detailed reply, and for the unexpected bonus of stumbling across evidence for my hypothesis. It’s late, I’ll reply to your other comments at a subsequent time … but just the above graphs are enough to falsify your argument that the January peak in global TSI was making the temperature peak in May. Instead, at the surface (which is the only sun the ocean knows, it knows nothing of TSI), the solar peak is in October …
But not only that. Rather than the temperature and the solar input varying in parallel as you and most other solar researchers assume, they are inversely related … and that fact alone is enough to falsify your whole argument.
Best regards, more later,
w.
W/m2 all over those two last SST graphs.
Thanks, fixed.
w.
Willis,
“Rather than the temperature and the solar input varying in parallel as you and most other solar researchers assume, they are inversely related..”
I don’t think that’s quite fair. Your residual in S4 equals the derivative of Nino3.4 SST:
http://climexp.knmi.nl/data/idnino5_daily_2000:2014dT_3_120day_low-pass_loess1a.png
so, the temperature and the integral of the solar input is varying in parallel (like it is supposed to when water is involved), temp is not the inverse of solar input.
I don’t understand this at all. The residual in S4 has nothing to do with the sea surface temperature. The underlying data of the absorbed solar is calculated from the CERES dataset with no reference to the Nino3.4 SST.
More to the point, it seems that you are claiming that the natural course of things is that when the surface insolation goes up, the ocean temperature goes down … how does that work?
w.
Willis, no
solar input = derivative of Nino3.4 SST
then
integral of solar input = Nino3.4 SST
Run the integral and you will see. I have done it for the all tropics (-30/30N) here:
virakkraft.com/CERES_EBAF_Surface_Net_Shortwave_Flux_Tropics.xlsx
lgl, it seems you missed my question, so I’ll ask it again.
Note that this inverse relationship is NOT true in much of the world. Elsewhere, when net surface insolation goes up, temperature goes up.
But in the Nino3.4 region, the opposite is true.
I say that this is because the temperature is regulating the clouds, which in turn are regulating the surface insolation. Warmer ocean ==> more clouds ==> less insolation.
w.
Willis
That’s not what I’m claiming, so your question is meaningless.
I’m claiming SST is increasing when the input is positive (anomaly > 0)
Willis Eschenbach September 24, 2015 at 1:22 pm
lgl September 24, 2015 at 3:48 pm
lgl, the data shows that when surface solar absorption increases the SST goes down. That’s a fact. You claim this is explainable because of the integral, or the differential … but regardless of your explanation, you obviously think that a surface cooling down when it receives MORE solar radiation is somehow the natural course of things.
Hence my question. It’s particularly relevant because in most parts of the world, the surface is reacting in the way we’d expect—when the solar input goes up, the temperature goes up.
But not in the Nino3.4 area. There, the situation is reversed. There, when the solar input goes up, the temperature goes down.
I say that this is because in the tropics the temperature is controlling the solar input, through the relationship
Warmer ocean ==> more clouds ==> less insolation, or alternately,
Cooler ocean ==> less clouds ==> more insolation.
You, on the other hand, seem to think that such a response (temperature goes down when forcing goes up) is actually just the normal course of things … but if so, why is it not true over the rest of the planet?
w.
Willis
lgl: “I’m claiming SST is increasing when the input is positive (anomaly > 0)”
Willis: “you obviously think that a surface cooling down when it receives MORE solar radiation”
Hm… wonder how your brain works …
Anyway, “the data shows that when surface solar absorption increases the SST goes down. That’s a fact.” Yes, like the integral of a sine wave is in counterphase with the sine wave half the time but still a result of the sine wave.
http://woodfortrees.org/plot/sine:10/last:300/normalise/plot/sine:10/last:300/integral/normalise
And, it’s also a fact that when the solar input is high the SST increase is also high, and solar input leads SST change.
http://virakkraft.com/Net-Solar-Nino34-SST-derivative.png
But this whole thing is a loop and it’s pointless discussing where the starting point of a loop is, so to close the loop you initiated:
Warmer ocean ==> more clouds ==> less insolation =>
=> Cooler ocean ==> less clouds ==> more insolation =>
=> Warmer ocean ==> ….
Wow, this is a great thread!! (OK, I am a little drunk) 🙂
Thank you Anthony for your wonderful site and Willis for his post ( I love your stuff) and David Douglass and Robert Knox for coming into the lions den and engaging. It’s all really interesting and educational stuff and although my brain struggles a bit I love it!
Thanks again to all!
Pretty rude response from Gosselin. He doesn’t actually pick one paper as you asked, instead throwing another four on the pile.
http://notrickszone.com/2015/09/24/4-new-papers-show-suns-impact-on-global-climate-german-scientists-sun-is-a-major-climate-factoR
I have no idea about the technicalities, but I’ve noticed the ‘solarians’ or whatever you call them tend to be dodgy after Willis’ articles. So they spend so much time to get the paper written and published, and then they can hardly be bothered to answer the criticisms.
I bet 2/3rds of these posts could have been eliminated and saved me from wasting a bunch of time had a “link” been provided to all the data.
Willis has a way of convoluting the data and presenting it in a very misleading mess to put it mildly to make it somehow try to support his wrong assertions when it comes to solar/climate relationships.
In addition Willis needs to look at prolonged periods of active or inactive solar activity.
Willis has convinced me that I am more correct then ever through his feeble attempts to show otherwise.
Salvatore Del Prete September 24, 2015 at 8:29 am Edit
This is great! I’m in the data convolution business … Salvatore, your complaint about my use of actual data to demonstrate relationships reminds me of the Homer Simpson quote:
w.
Willis, I am just saying to you what you have said to those of us who support a solar/climate relationship which is we are in the business of using and presenting convoluted data.
The feeling is mutual and only time will tell the truth.
That’s bovine waste products without a quote to back it up. Are you too stupid or too arrogant to follow a polite request, viz:
Find a quote where I said that someone is “in the business of using and presenting convoluted data” and we can discuss it. Until then, you’re just slinging mud, which proves you’re out of scientific ammunition.
I will not allow you put words into my mouth that I never said, that’s underhanded and deceptive.
w.
Having gotten some sleep, let me return now to the responses that Drs. Douglass and Knox have kindly provided to some of my objections to their work. My thanks again to them for responding, it is unusual and noteworthy in a time when many scientists refuse to defend their work in public.
To the responses, they’ve given the quotation of my words first, my thanks to them for clearly identifying their objections:
So indeed, they were previously reported by the authors themselves, as I said …
I fear you have not answered my objection, which was the lack of a a citation. I’m sure you know where they were “previously reported”, since you wrote it. And they may indeed be in your list of references. However, telling me that the information is only two clicks away doesn’t help—the entire web is a couple of clicks away.
The problem is that without a citation or link of some kind, there’s no way to know who or what was “previously reported”.
The question is not “where is the data”. The question is why you didn’t analyze it.
So indeed, the data was being stubborn and uncooperative, as I said …
More to the point, I’m sorry, but saying in effect “If I included all the data I’d have to discuss it more“ is not a sufficient reason to not analyze a quarter of the data.
And that is ten times as true when no reason is given in the paper for not analyzing such a large chunk of the data … you gotta admit, the optics of that are horrible.
Yes, I had read and understood that. Your argument is that a boxcar smooth is better than removing the climatology because it is smoother. This allows you to use it to split the data into a high frequency-low frequency pair which you can analyze separately. Fair enough.
However, this splitting of the dataset is for a very different purpose than the purpose behind removing the climatology. So I see the two methods as being complementary to each other, as they are used for different purposes.
In terms of the precision, the climatology removal is more precise. I say this because it doesn’t mess with any longer periods. As the periodograms show, the boxcar filter not only removes the 1-year signal, it also reduces cycles up to about 6 years in length.
I did not catch that in the intro, my bad. But that wasn’t my main point. I’m sorry for my lack of clarity. What I meant was, where are the other segments in the actual analysis? What I was interested in was the fact that some of these segments would likely be in the missing quarter of the data … so why weren’t they analyzed?
And indeed, you say that segment #8 is in the missing data … but as you say, you didn’t include it because it’s too hard to explain. Got it.
Fourier analysis is now “meaningless”, but only for periods over one year? How on earth did I miss the memo? Heck, it’s even got a peak at 3 years … is that “meaningless” as well? Because that’s the very cycle that you claim is so important …
I fear you have a fundamental misunderstanding of the nature of Fourier analysis. It doesn’t “average” frequencies, ever. It just shows the strengths of the individual cycles of different period lengths (or frequencies). In the actual data the various cycles can be totally intermingled, going into and out of phase … doesn’t matter. The Fourier analysis separates them all out, and shows the individual strengths at each period length. For example, if there is a cycle of period 3 years and one of 5 years, you’ll get a beat frequency in the actual data. But that doesn’t matter to the Fourier analysis. It shows it as 3 and 5 years, and more to the current point, it never shows up as an average period of 4 years.
So those cycles are assuredly real. However, we do not know whether they are persistent because the data is only 32 years long, far too short to look at most frequencies.
To sum up:
• As I showed in the first half of my response to the authors above, when the absorbed solar radiation at the surface goes UP in the Nino3.4 Region, the surface temperature goes DOWN. This fact alone is sufficient to falsify their entire argument.
• The authors claim that the TSI solar input to the Nino3.4 Region peaks in January and is at a minimum in July. This is not true at the TSI, and it is even less true at the surface, which is what the ocean is actually responding to. Again this is very strong evidence against their argument.
Finally, even if those were not true, I’m still very uneasy about their method. They are looking for periods when there is either a 2-year or a 3-year cycle … but they’ve ignored all other possible cycle lengths. In addition, they are declaring “phase-locked sections” based on less than two cycles of data … sketchy. It seems to me that they are doing what is functionally a very simplified and clumsy version of wavelet analysis. If they’d actually done wavelet analysis it could have been interesting … if their theory hadn’t already been falsified by the two issues listed just above.
w.
So robustness is lacking. If an effect is found in a manipulated data set (versus direct observation), authors are ethically bound to determine the robustness of that effect when in observation mode. And effect may be found, but it may not mean anything to boots on the ground conditions. Robustness is meaningful across observations and data manipulation.
Willis
I am quite familiar with the CERES web site but can not find the TSI data in the first panel of your figure S4.
Please tell me how to download this data.
Thanks
David Douglass
David, there are two datasets at the CERES website. One is the TOA data, and one is a derived (although very accurate) set of surface data. Hang on …
OK, go here, and select the “Solar Flux” field. There’s only one subfield, called “Incoming Solar Flux”. Click on the “Get Data” button at the bottom.
You likely know the rest, but the data is in the form of an NetCDF file. It contains the observations in an array of 180 x 360 x 168 cells, which is 180° latitude by 360° longitude by 168 months of data.
Best regards,
w.
Now I might have missed it, but as far as I know, to date not one person has taken me up on my offer to analyze their one best solar study if they sent me TWO LINKS, one to the paper and one to the data. I mean, this is a chance to show me up, a chance once and for all to show I’m wrong … but not one person came to the party.
But there is some good news. Many folks have peanut buttered over this lacuna by concealing it beneath a thick layer of vigorous and vociferous abuse directed at my many noble, amazing, and even awe-inspiring qualities … so at least there’s been something to amuse us all while we’re waiting.
w.
Willis:
I’m still trying to find you a paper with data. Wow, it’s really really hard.
For example the Soon, Connolly & Connolly paper looks intriguing, they go through great lengths to explain all the problems with TSI and temperature reconstructions, which toggles my “they’ve read Feynman” bit.
However, where the heck is the data? They did their own temperature reconstruction using rural-only stations, but don’t link to it. Googling for Soon I can’t even find his academic home page, the first two Google pages are filled with smears (as is Wikipedia). As far as the TSI reconstruction, there are so many and Google can’t find the one they use in the paper.
I look in the references section, and it’s ALL papers. No data. It’s as if the data is an afterthought.
coming from the software world, this is just stupid. You just upload your software to github. Handy URL right there. Done.
Peter
https://nextgrandminimum.files.wordpress.com/2015/09/soonconnollyconnolly15-sep4-esrunformattedpreprint.pdf
Peter, many thanks for your perseverance. The lack of data is no surprise to me. It’s one of the things that drove me to only accept papers for analysis if there is a clear link to the data. People would give me links to half a dozen papers with a supercilious sneer, claiming the papers clearly established the sunspot cycle/climate connection … and after a pile of searching I’d find out that not one of them had data to support their claims.
The surprise to me is how many people buy into the data-free papers, and quote them as though the peer-review were enough for them. You’d think skeptics would be more … well … skeptical. I mean, most every skeptic knows that peer-review is meaningless, but as soon as some new solar claim surfaces, the true believers snap it up and start retailing it at a rate of knots as though it were established scientific fact …
And you are correct that this is all easy stuff. You just upload your data and code to some convenient location. I use Dropbox, so anyone can easily find the data (and code as appropriate) for my work. And for many datasets, you don’t need to upload them, just provide a link to the one(s) that you used.
Anyhow, your dedication is much appreciated. I just wish that the people claiming over and over that “it’s the sun, stupid” would do the work you’re doing, so that they can see just how thin on the ground actual data is … but then, I suppose that’s exactly why they don’t do it …
My best regards to you,
w.
Willis,
A new blog post here (http://edberry.com/blog/ed-berry/new-study-sun-not-co2-causes-climate-change/) cites a paper by here (http://globalwarmingsolved.com/data_files/SCC2015_preprint.pdf) with data files provided here (http://globalwarmingsolved.com/data_files/SCC2015-SI.zip).
When you have time, would you mind doing us the honor of hitting it with a sledgehammer a couple of times and seeing if it breaks? ^_^
Thanks in advance!
Sorry, the paper is by Willie Soon, Ronan Connolly, and Michael Connolly. An EDIT button would be nice, wouldn’t it…?
Aaaaand, it looks like maybe I found the data everyone was looking for….?
Thanks, Smokey. There are two huge problems with the study.
The first is that they are using the old sunspot data. This data incorrectly says that there was an increase in solar strength over the last couple centuries. Using that data automatically invalidates any conclusions they might draw from it.
The second is that it seems that they can’t find any correlation with any of the standard temperature datasets, so instead they made up their own special dataset. This is an amalgam of the temperature records of “rural” stations in China, the US, Ireland, and the Arctic …
In addition to the fact that their reconstruction doesn’t look much like any of the standard temperature records of the same period, I have a huge problem with the procedure. There are literally dozens of areas of the planet that they could have chosen for their reconstruction, giving untold possible reconstructions. So there are quite probably one or more reconstructions that one could create that would give a good correlation with sunspots, or CO2, or the price of US postage stamps, or the increase in world population over the time … so what?
As a result of both of those problems, I wouldn’t even have a clue how to calculate the statistical significance of any results one might get from comparing a purpose-built temperature reconstruction with an incorrect sunspot dataset … but it’s not gonna be pretty, and there’s no way it could be significant. Bad data in, bad data out is the iron-clad rule.
My thanks to you, and to the authors of the study for making their data available.
w.
Hi Willis,
Somebody mentioned you had just commented on our recent Earth-Science Reviews paper (http://www.sciencedirect.com/science/article/pii/S0012825215300349) here. Unfortunately ESR is a paywalled journal, but as Smokey notes above, I’ve put a copy of a pre-print here (http://globalwarmingsolved.com/data_files/SCC2015_preprint.pdf) and a copy of the SI here (http://globalwarmingsolved.com/data_files/SCC2015-SI.zip).
You say there are two huge problems with our paper. I suspect this may actually be due to a misreading of what we said? I appreciate the paper is quite long, so it’s possible this has led to some confusion…
Willis: “The first is that they are using the old sunspot data. This data incorrectly says that there was an increase in solar strength over the last couple centuries. Using that data automatically invalidates any conclusions they might draw from it.”
Actually, this is not correct. First off, we didn’t actually use sunspot numbers for our analysis! To be honest, I’m not sure where you got that impression, as we were quite critical of studies which placed too much reliance on sunspot numbers (whichever version) as an exact proxy for solar variability. But, I do appreciate it is a long paper, and quite a lot of other studies have relied on sunspot numbers, so it is possible you just assumed this was another example.
Instead, we used some of the different published TSI reconstructions which were available in the literature, i.e., some of those in Figure 8. It is correct that most of these reconstructions use sunspot numbers as one of their solar proxies, and since all of these reconstructions predate Clette et al., 2014’s recent update (which AFAIK was only published this summer?), the versions of the sunspot numbers that they would have used would have been one of the other datasets. But, while most of the reconstructions rely quite heavily on sunspot numbers, most of them use more than one solar proxy. Indeed, the updated Hoyt & Schatten, 1993 reconstruction (which we discuss in detail in Section 5) actually consists of seven different solar proxies – see Section 2.2.4.
Moreover, in our discussion of the sunspot number records (Figure 4), we consider all three of the main versions, i.e., the original Wolf a.k.a. Zurich a.k.a. International numbers (see Figure 4a); Hoyt & Schatten’s “Group sunspot number” (see Figure 4b) and Clette et al., 2014’s recent update (see Figure 4c). Did you read Section 2.2.1 where we compare and contrast all three of these datasets as well as several other solar proxies aside from the sunspot numbers?
At any rate, although the Wang et al., 2005 TSI reconstruction – which the latest IPCC Global Climate Model hindcasts (“CMIP5”) mostly relied on – is very closely related to the Hoyt & Schatten Group sunspot numbers (see Figure 9), and therefore our comparison of our Northern Hemisphere composite to the CMIP5 hindcasts in Figure 25 is indirectly based on a comparison with what you call “the old sunspot data”, this was the IPCC’s choice, not ours! We actually criticised this choice (in both Sections 2.2.4 and 4.5), and argued that climate modellers should consider a much wider range of TSI models for their hindcasts.
We argue that, at a minimum, they should have considered all 8 of the various TSI reconstructions in Figure 8 that have been published recently in the literature, and not just the four “low solar variability” reconstructions on the right hand side of Figure 8.
Do you not agree yourself?
Moreover, a major point that we stressed in Section 2.2 is the fact that, while initially most of the solar proxies seem to imply fairly similar trends, when you look in detail at each of the solar proxies, there are quite a few differences between each of them – sometimes subtle, but sometimes quite pronounced.
In your comment, you note one such difference, i.e., the differences between Clette et al.’s recent reanalysis of the sunspot number data (Figure 4c) and the two other more commonly used sunspot number datasets – Hoyt & Schatten’s group sunspot numbers (which the Wang et al., 2005 TSI reconstruction recommended by CMIP5 relies on) and the original Wolf/Zurich/International dataset (Figures 4b and 4a respectively). But, as we illustrate graphically in Figures 4 and 5, there are quite a lot of plausible solar proxies, each implying slightly different trends and peaks.
You seem to be arguing that researchers should not rely on the datasets in Figure 4a or 4b, but rather rely on the new one in Figure 4c. Is that correct?
If so, we’ve actually gone further and recommended that instead of simply relying on one of the ~20 solar proxies in Figures 4 and 5, we should be recognising the subtle and not-so-subtle differences between all of the different plausible solar proxies.
Do you not agree?
Willis: “The second is that it seems that they can’t find any correlation with any of the standard temperature datasets, so instead they made up their own special dataset”
Actually, this is not correct either, and seems to miss the point of the second half of the paper. As we summarised in the abstract and explained in the Introduction (Section 1), there were two main goals of this paper:
1. To try and accurately and fairly review the extensive debate over solar variability.
2. To try and minimise the urbanization bias problem associated with the standard global/hemispheric temperature reconstructions, by developing a new estimate of Northern Hemisphere surface air temperature trends since 1881.
So, we did NOT develop this dataset because we “…can’t find any correlation with any of the standard temperature datasets” – we specifically developed the dataset a priori to try and minimise the urbanization bias problem. Not only that, but in Section 4.1, we carried out a detailed discussion of how our new estimate compares and contrasts with the previous land-based estimates, i.e., the ones which included both rural and urban stations.
Willis: “There are literally dozens of areas of the planet that they could have chosen for their reconstruction, giving untold possible reconstructions. “
Initially, this seems like a fairly obvious and valid criticism. Before I had analysed the GHCN dataset (in considerable detail!!!), I would have thought the same as you… However, remarkably when you look carefully at the GHCN dataset, there is a serious shortage of regions with a large number of fully rural stations with reasonably long and complete station records.
I’m not sure how familiar you are with the GHCN dataset, but have you looked in detail at the spatial and temporal distribution of the fully rural stations? Most of the fully rural stations in the GHCN actually only have data for the 1951-1990 period! There are quite a lot of regions which have reasonably long and complete station records… but, these stations are generally not fully rural ones!!!!
We tried to identify as many regions as we could for which we there was enough data to construct a reasonably reliable long-term rural estimate. You would (rightly!) expect that there should be “dozens” of such areas, but the reality is that there currently is not!
The U.S. is of course a region with a high density of fairly long and complete rural stations… thanks to the USHCN, and fortunately thanks to the Surfacestations project (which I know you were involved in!), we also have some idea on their microclimates. But, for the rest of the world, there is a surprising shortage of fully rural stations with long & complete records…
Willis: “In addition to the fact that their reconstruction doesn’t look much like any of the standard temperature records of the same period”
Actually, as we discuss in Section 4, while our new Northern Hemisphere temperature reconstruction implies the previous land-based estimates have slightly overestimated the 1980s-2000s warming and underestimated the 1950s-1970s cooling, it is actually quite consistent with SST (Section 4.2), glacier length-derived estimates (Section 4.3) and tree ring-derived estimates (Section 4.4). Our new reconstruction doesn’t show a very good fit to the CMIP5 hindcasts (whether using their “Natural” and/or “Anthropogenic” forcings) – see Section 4.5. However, it does show some evidence which is consistent with an influence from stratospheric volcanic cooling (Section 4.6).
– Ronan
P.S. Sorry for the long reply. In the words of Blaise Pascal, “I have made this [comment] longer than usual because I have not had time to make it shorter”!
Ronan Connolly October 3, 2015 at 4:49 pm Edit
I don’t understand this comment at all. You say below that all of the reconstructions used the sunspot numbers, and you agree that in fact they used the old sunspot numbers:
The TSI reconstructions all used the old sunspot numbers. You used the TSI reconstructions. Ergo, your analysis depends on the old sunspot numbers. What am I missing here?
You did discuss and consider the three main versions (Wolf, Hoyt/Schatten, Clette). However, in the event you used the old ones, so I’m not sure what good discussing and considering them did.
As I read this, you are saying that indeed you DID use the old sunspot numbers, but only because it was the IPCC’s choice. Fair enough.
Mmmm … my preference would be to study the eight individual TSI reconstructions first, and see if I agreed with ANY of them. Next, I’d be clear and say that none of them might be right … and that at a maximum, one of them might be right. Given that, if you consider them all, the same would be true … bad odds.
Next, when you consider all 8 of them, you need to use the Bonferroni correction on your results … which means that if you think that a p-value less than 0.05 is necessary for significance, when you consider eight of them you need to find something with a p-value of less than 0.05 / 8 = 0.006 … which is a tough ask.
My point exactly. All of them can’t be right.
Yes, I agree.
Not in the slightest. As I pointed out, and as you agree, there are a number of differences in all of the ~ 20 solar proxies. This means that at most only one is right, and 19 are wrong. The idea that if we just consider all of them we’ll be closer to the truth seems … well … optimistic.
It sounds to me like you are saying that because the regular datasets contain some unknown measure of distortion due to the UHI (the “urban bias problem”), that you can’t find the solar signal in them.
How is that different from what I said, viz:
I take a lot of grief because I make obvious assumptions. In this case, I assume that if you’d been able to find the desired correlations with the datasets containing UHI you would have used them and not bothered with making your own … but you didn’t use them.
I’m sorry, but that makes no sense. You’ve used four of what you call “regions”, viz, US, China, the Arctic, and Ireland. Why on earth would you limit yourself to “regions”? Heck, three of the four “regions” are countries … are there no rural stations in Canada? The sensible way would be to use all available rural stations. Instead, you use restrictive regions like “Ireland”. As near as I can tell, you have one Irish station with data before 1950 … WUWT?
But regardless of the choice of regions, you’ve made it impossible to compute any statistics on your results, because you could have included other stations in your reconstruction and you did not do so.
Finally, it doesn’t appear that you have any ex-ante criteria for station selection. For example, you say:
Despite that you have a whole bunch of non-US stations. You also say:
I fear I couldn’t find the “required characteristics for a particular subset”, but the idea that there are different requirements for different subsets is worrisome. Also worrisome is the idea that FIRST you selected the regions and then you only considered stations within the regions.
In addition, despite using your own new corrections to other stations, you don’t correct stations in the Arctic for UHI …
So you’ve ended up with some adjusted stations done by your new and “improved” method (which it may be) and some stations adjusted by other methods, and some stations not adjusted at all.
Wait a minute. Your claim is that your instrument-based reconstruction is supported by the fact that it is “quite consistent” with the tree-ring and glacier-length reconstruction? Doesn’t it usually work the other way around, with instruments used to verify proxies? Color me unimpressed.
Look, there are a couple of independently developed instrumental datasets for historical temperature (GHCN-based, and Berkeley Earth). They look a lot like each other. And, they both look a lot like the satellite records. Your reconstruction doesn’t look like either the satellite records OR the station based records. Given that, the fact that it looks like the tree-rings doesn’t mean much …
Since the CMIP hindcasts are tuned to the accepted historical temperature record, they would not fit your record, so that is no surprise.
And if it shows “some evidence which is consistent with” a volcanic influence, that means that it shows some evidence which is NOT consistent with a volcanic influence … sorry, but if you’ve read my work regarding volcanoes, I don’t find “some evidence” to be anything like persuasive. Most people doing that analysis forget that there is a 50/50 chance of temperatures going down after a volcano … so “some evidence” is meaningless.
Not a problem. Thank you for taking the time and trouble to answer in detail, it’s the only way to make your meaning clear. Comments should be as short as necessary and no shorter.
Regarding this whole thing, I would say that yours is a portmanteau study. You have simultaneously attempted an overview of TSI reconstructions, proposed an entirely new temperature reconstruction, and looked for a connection between the two. As a result, I fear you’ve ended up with none of them being firmly established.
And in the end, the best we can say is that using TSI reconstructions based on outdated sunspot numbers, you’ve found a correlation with your just-born and completely unverified historical temperature estimation … I couldn’t even begin to guess at the statistical significance of such a finding, but I find it completely unconvincing as regards the solar influence on the climate.
Best regards, and thanks again for your clear comment.
w.
Willis & Ronan,
Thanks for the back & forth on this. Having looked through both sets of your comments, the paper & the data set, I must tentatively admit that I come down in favor of the new paper… while also admitting that it is by NO means a slam-dunk “proof.”
Even though this paper does not seem to be the “DNA evidence” which proves solar culpability in re: the observed historical climate (at least, not beyond all reasonable doubt) I still feel that the datasets involved are derived in good faith and according to valid, logical premises; the conclusions based upon their comparison are thus both interesting and compelling, if not nearly the last word on the subject. If more definitive data, correlations and conclusions along the same lines can be uncovered, I doubt anyone will really be all that surprised in retrospect.
THAT SAID: I think definitive confirmation/refutation of the solar cycle upon decadal/centennial climate will have to wait until we can, e.g., look back 100+ years from now at a much longer satellite record of both temperatures AND solar activity — a record which will hopefully be nearly unassailable in terms of continuity, accuracy and coverage, the lack of such records today having contributed mightily to our current predicament in terms of “which set shall we use?” or “should we create a new set?” and so forth — which will hopefully show clearly that these fellows were on a/an [right/wrong/irrelevant] (circle one option) track.
Until then, we’ll just have to wait & see. Thanks again, both of you, for your time!
Hi Willis,
Ah, thanks for your response – I think we may have been talking at cross purposes, as I actually agree with many of the points you make, and in most cases we made similar points in the paper!
Perhaps it would help to understand a bit of the background behind why Michael (my father) and I did this collaborative study with Willie Soon. For the last few years, Michael and I have been studying the various attempts which have been made to quantify global temperature trends. In early 2014, we submitted several papers discussing our analysis for open peer review at the Open Peer Review Journal (a website we set up as an ongoing attempt to try and open up the peer review process): http://oprj.net/category/articles/climate-science
I know you are familiar with our “Global temperature changes of the last millennium” review of the various millennial temperature proxy reconstructions, since you made some very useful and helpful comments on it. But, have you had a chance to read any of our papers analysing the various weather station-based reconstructions yet?
If not, then one of our main findings was that the insidious nature of the urbanization bias problem has been seriously underestimated by the various groups who developed what you call “the standard temperature datasets” – see in particular, “Urbanization bias I. Is it a negligible problem for global temperature estimates?”: http://oprj.net/articles/climate-science/28
However, our approach to science has always been heavily driven by a sense of curiousity, and so, to us saying “the data used by the previous studies is problematic” is very unsatisfying – we want to know, “well what is the data saying then?”. So, in our “Urbanization bias III. Estimating the extent of bias in the Historical Climatology Network datasets” study (http://oprj.net/articles/climate-science/34), we decided to look in detail at the various monthly temperature datasets (of which the GHCN was – at the time – the most commonly used) and see if (aware of the insidious nature of the urbanization bias problem) we could do a better job…
As we discuss in that paper (and as I mentioned in my comment yesterday), for the U.S., we were able to identify enough fully rural stations with reasonably long and complete station records to construct a regional estimate for U.S. temperature trends since the late 19th century. However, remarkably, for the rest of the world there were very few stations that met those requirements – which initially we had assumed (like you!) would have been fairly basic requirements. We did find a handful of fully rural stations which met those requirements in Europe, but each of these stations implied slightly different trends, and there was evidence that several of them may had been affected by non-climatic biases associated with station moves, changes in microclimate, instrumentation, etc.
Unfortunately, aside from the USHCN component, the GHCN dataset doesn’t include any station histories. None of the other datasets (i.e., Berkeley Earth, ISTI, CRU, etc) do either! So, we concluded that without access to the station histories for these stations, it was very difficult to determine which (if any!) of the station records was most representative of the long-term rural temperature trends for the region.
One of the main recommendations we made in that paper (and some of our others) was that effort should be made into compiling station histories for the station records, so that the non-climatic biases which are associated with long station records could be reduced. [As an aside, we found that there are serious problems with the automated homogenization algorithms that are currently being used – in the absence of actual station histories – to account for non-climatic biases ]
We also suggested that, if one of the new datasets (ISTI or Berkeley Earth) were used, then it was likely that more stations meeting those requirements could be identified. However, while the compilers of the GHCN dataset had put quite a bit of effort into identifying how urbanized the stations were, this has not (yet!) been done for the ISTI or Berkeley Earth datasets – although Wickham et al. did carry out a preliminary categorisation of the Berkeley Earth stations (I gather with assistance from Steve Mosher).
By the end of 2014, in discussion with some of the ISTI group, we had started analysing the ISTI dataset with a view to sorting the stations according to their degree of urbanization. This is ongoing work, but (like you) none of our climate research is funded, and so we can only carry it out in our spare time. 🙁
This comment is already quite long, so I’ll continue in a separate comment! …
Ronan Connolly October 4, 2015 at 12:56 pm
Thanks for continuing the discussion, Ronan, much appreciated.
Yes, it’s a problem. Is it “seriously underestimated”? I think it’s underestimated, but I don’t think anyone has a good handle on the size of it.
Oooh, you are omitting data from your study because it doesn’t do what you expect … this is egregious data snooping.
Agreed
Agreed.
Here’s my question, which all of this hasn’t answered … why didn’t you a) put out ex-ante criteria for which stations you would use, and then b) use EACH AND EVERY station that fit your ex-ante criteria, whether or not they did what you expected it to do.
Onwards to your next comment,
w.
…At any rate, at this point Willie contacted us. As you probably know, he has been carrying out a lot of research into evaluating the role of solar variability in climate change, and in particular into the possible influence of solar variability on recent global (and regional) temperature trends. However, until he read our papers, he had been assuming (like you!) that the standard temperature datasets were reasonably reliable. If the standard temperature datasets are significantly affected by urbanization bias (and other non-climatic biases), as our work suggests, then this could could have implications for his analysis. So, he wanted to know what dataset we recommended he should be using instead.
We explained that, aside from the U.S. component, the GHCN dataset was surprisingly limited and problematic, and that our analysis of the ISTI dataset was still ongoing. However, after some discussion, we decided that if we looked in detail at the other regions on a case-by-case basis, there were actually a few more regions where we could carry out a cautious and caveated “best current guess” estimate, i.e., China, Ireland and the Arctic.
This was what led to one half of our Earth-Science Reviews paper, i.e., the construction of a new mostly rural Northern Hemisphere temperature reconstruction from 1881-2014.
Because Michael & I live in Ireland and are only ~4 hours drive from the main rural station in Ireland (i.e., Valentia Observatory), we were able to contact the station owners directly and they provided us with a detailed station history and useful parallel measurements associated with 2 of the main station history events. In other words, for this station, we were able to implement the recommendation we made in our 2014 papers. It would be nice to extend this analysis to include some of the other long rural records in Europe, and in Section 3.3, we recommend this should be done by the collection of similar station history information. We think something like what you guys did with Surfacestations could be a useful way to do this, but at any rate, we think this should be a top priority for improving the reliability of the temperature datasets…
For the remaining two components (China & Arctic), we had to take a number of assumptions and provide cautionary caveats (for instance our concern about some UHI in the Arctic). However, in the interests of providing greater hemispheric coverage, we decided it would still be of interest.
We spent a long time seeing if there were any other regions which we could also include with some measure of confidence in their reliability. For instance, you mentioned Canada – we did actually include all of the available Canadian Arctic stations in our Arctic composite, but we also considered expanding our Arctic composite to include sub-Arctic Canada (and similarly sub-Arctic Russia). However, as it is, of the four components, the Arctic component is the one we probably have the least confidence in… even though our estimate is very similar to the regional Arctic estimates derived from “the standard temperature datasets”. That is, the groups compiling the standard temperature datasets apparently would have more confidence in our Arctic component than we do!
We are looking to see if we can improve on this Arctic component for future work, but we will probably not be using the GHCN for this.
Although there are (as you imply) plenty of other regions in the Northern Hemisphere which have stations with fairly long and complete station records, these are urbanized stations!
For us, the biggest gap in our Northern Hemisphere sampling is probably for the tropics – a point we stress in Section 4. But, again this is because there is a severe shortage of fully rural tropical stations with long records in the GHCN.
To summarise, the main emphasis we took in our approach was quality over quantity. While the groups who developed the standard temperature datasets seem to feel it’s ok to ‘throw everything into the pot’ and see what comes out, we feel that this is not appropriate. What we have found is that, despite the claims of several groups, the underlying data is problematic. We think it is important to recognise this and treat the data carefully and be explicit about our caveats and assumptions.
For our earlier work, we applied what we thought should have been fairly basic minimum requirements… but essentially the only part of the GHCN dataset which met those requirements was the U.S.! For this reason, in this study, we tried to lessen these restrictions a bit to try and increase our station coverage. This gave us two new regions, i.e., China and the Arctic, but we stressed in the paper that these new estimates still need to be treated cautiously (see the discussions in Section 3).
We also followed up on our own recommendation that station histories could be used to help overcome some of the challenges, by tracking down the information for our local long-term fully rural station. This opened up a third region, Ireland. Yes, it’s a small country, and it would be nice if this analysis could be extended to include some of the other long-term fully rural stations in Europe. But, remember, all of this work is self-funded! We haven’t yet managed to track down the station histories for these other stations.
It’s true if we completely dropped our basic requirements for the data (as the other groups have), then we would get basically the same result as them. Indeed, we did this in Section 4.1 (see Figures 20 and 21). It’s kind of like how if I construct a temperature proxy reconstruction using bristlecone pines, Yamal and upside-down Tiljander lake sediments, I can create a “hockey stick” too… 😉
But, should we really abandon our basic requirements just so that we can be “part of the gang”? We don’t think so. We’re more interested in trying to figure out what the data is really telling us, even if it means going through it piecemeal… What do you think, yourself?
Does that background give you a better insight into our approach?
Also, does it explain why this assumption you made is wrong:
Willis: “I take a lot of grief because I make obvious assumptions. In this case, I assume that if you’d been able to find the desired correlations with the datasets containing UHI you would have used them and not bothered with making your own … but you didn’t use them.”
Michael & I weren’t even considering solar variability when we started this – our main goal was (and still is) simply to do the best we can with the currently available data to minimise UHI and other non-climatic biases, etc.
Willie, on the other hand, had been using the standard temperature datasets all along, and probably would have continued to do so. But, once he realised that there were potentially serious problems with them, he decided this wasn’t good enough!
I guess this brings us to the other half of our paper, i.e., the solar variability debate, but this comment is also very long, so I’ll continue in yet another comment…
Ronan Connolly October 4, 2015 at 12:57 pm
As I said above, if the solar analysis had proven fruitful with the standard temperature datasets, Willie wouldn’t have needed a new dataset to compare with. You objected strongly to this characterization, but it appears to be true.
I still don’t understand why you are messing with “regions”. Why not just include all valid records, whether they fit into a “region” or not?
Same objection. You are stuck on the idea of “regions”, and as a result you are not using all of the data.
I’m sorry, but I simply do not believe that the only valid stations for your research are in the US, China, Ireland, and the Arctic. That makes no sense at all
Again, same objection. If there is a “severe shortage”, that’s all the more reason to use the few that remain.
But you haven’t done that. You haven’t established ex-ante criteria and only used what fit. Instead, you appear to have special criteria for each of your different “regions”.
Never mind about the “regions”. You’ve blinded yourself with your insistence on “regions”.
I think that your claim that you have “opened up a third region” reflects a total misunderstanding of what you are about.
It explains what you are doing wrong …
My point was that if Willie had found the correlations he seeks in the usual datasets, he wouldn’t need a new dataset. In that, I was correct. However, as you point out, I was 100% wrong that that this was the reason for creating your dataset, because clearly you wanted to create a dataset that was free of UHI.
My final question would be this: the satellite temperature datasets track the surface temperature datasets quite closely, although they differ somewhat in the overall trends. And presumably the satellite datasets are NOT affected by UHI.
So why don’t your results track the satellite data, since you claim your results are also free of UHI?
w.
…With regards to the solar variability debate, in some senses, Michael & I have a fairly similar perspective to you. When we first started reading the literature and analysing the data on this topic, we found (like you!) that most of the “It’s the Sun, stupid” studies were one-sided and often relied on cherry-picked and/or incomplete analysis.
However – and perhaps we diverge from you here? – we also found that the same applied to most of the “It’s NOT the Sun, stupid” studies!!!
We found that – on both sides – a lot of the studies seem more interested in researchers ‘defending their turf’ and ‘affirming the consequent’, and less interested in trying to genuinely establish what role (if any) the Sun has played…
Importantly, as we discussed on our blogpost assessing the IPCC process (http://globalwarmingsolved.com/2013/11/what-does-the-ipcc-say/), the CMIP5 hindcasts which the IPCC relied on for their “Detection and Attribution” studies seemed to have taken a very one-sided approach to modelling “natural forcings” by only focusing on the low solar variability TSI reconstructions – see in particular Section 3 of that blog post (i.e., the ‘How was this “extremeley likely that…” conclusion reached?’ section).
So, when we were discussing with Willie the various TSI reconstructions and what we actually know (and don’t know!) about how solar activity has varied since the 19th century, we all decided that it was time that somebody did a balanced and comprehensive review of the literature (and data) presenting all perspectives.
We tried our best to be as non-judgemental as we could in our review (Section 2), and wherever we had strong opinions on a particular aspect, to flag them as such to the reader. Of course, we have to keep reminding ourselves that if you think you are unaffected by confirmation bias, then it’s quite likely that you are affected! 😉
But, we tried to do our best…
At the end of our review, our main conclusion was that… this is actually a very tricky problem, and we still don’t really know for sure!!! 🙁
See Section 2.4 for our summary of the main points of general agreement and the main points of debate currently in the literature.
So I actually completely agree with you when you say,
Willis: “Mmmm … my preference would be to study the eight individual TSI reconstructions first, and see if I agreed with ANY of them. Next, I’d be clear and say that none of them might be right … and that at a maximum, one of them might be right. Given that, if you consider them all, the same would be true … bad odds.”
and
Willis: “Not in the slightest. As I pointed out, and as you agree, there are a number of differences in all of the ~ 20 solar proxies. This means that at most only one is right, and 19 are wrong. The idea that if we just consider all of them we’ll be closer to the truth seems … well … optimistic.”
In our opinion, neither the “It’s the Sun, stupid” nor the “It’s NOT the Sun, stupid” groups have satisfactorily resolved the issue.
However, there are several conclusions we think we can confidently draw from our analysis:
1. The CMIP5 hindcasts (which formed the basis for the latest IPCC report’s main conclusions) only considered a small (and one-sided) subset of the currently available TSI reconstructions (see Figure 8). Therefore, their treatment of “natural forcings” in their detection and attribution studies is very limited, and as such, should be treated with considerable caution.
2. While the CMIP5 hindcasts with “anthropogenic + natural forcings” seemed to do a reasonable job of hindcasting the standard temperature datasets, they seem to have done a rather poor job of hindcasting our new mostly rural dataset.
We can say that the standard temperature datasets are – at least to some extent – significantly affected by urbanization bias, e.g., see the evidence of urbanization bias in the China data over the 1951-1990 period (Figure 10), since only 1/4 of the Northern Hemisphere stations (in the GHCN at least) are fully rural. On the other hand, our new Northern Hemisphere reconstruction is mostly rural, with 86.4% of the stations being fully rural… and it is constructed from a large chunk of the fully rural NH stations that are available in the GHCN (42.2%) – see Table 6.
Also, there are problems with the automated Menne & Williams homogenization algorithm which the groups using the homogenized GHCN dataset relied on, e.g., see Sections 3.1 and 3.3.
Therefore, even if you dispute the reliability of our new Northern Hemisphere reconstruction (and we have stressed that this is just our current best attempt to deal with the problems!), there are also problems with the “standard temperature datasets”. So, the apparent goodness of the CMIP5 fit used by the IPCC to conclude that “it’s NOT the Sun, stupid” is no longer valid.
It also means that, if “it’s still not the Sun, stupid”, then the CMIP5 models are leaving out at least one key driver that seems to have played a major role on temperature trends since the 19th century – and it doesn’t seem to be proportional to either GHG or man-made aerosols (i.e., the CMIP5 models’ “anthropogenic forcings”). If it’s not the Sun, then identifying this apparently “missing” factor should become a top research priority. It is plausible that this could be a “natural forcing”. This would concur with the arguments people like Roy Spencer have been making for years, i.e., climate scientists should be actively encouraged to investigate the role natural variability has played on recent climate change.
3. One of the current TSI reconstructions which the CMIP5 modellers ignored (i.e., ACRIM’s update of the Hoyt & Schatten, 1993 reconstruction) does seem to show a remarkable fit to our new Northern Hemisphere composite… and also to each of our four regional components (if you still dispute the representativeness of our composite). So, if the updated Hoyt & Schatten, 1993 TSI reconstruction is at all accurate, then this suggests that solar variability has been a primary driver of temperature trends since at least 1881 for much of the Northern Hemisphere – regardless of whether you think it was just limited to the four specific regions in Figure 19, or for the entire hemisphere.
This suggests that the “it’s the Sun, stupid” argument is at least a plausible one, and more research should be carried out into establishing which TSI reconstructions (if any) are most representative of the actual TSI trends. If it transpires that none of them are reliable, then more research should be carried out into developing new TSI reconstructions.
Do you agree that those specific conclusions are reasonable ones to draw from our analysis?
If so, then I think we’re probably mostly in agreement on this, after all…
On a separate issue, while I think Leif Svalgaard (who often comments here) and the rest of the Clette et al. group have put a lot of valuable work into their new version of the sunspot record, and I acknowledge that Leif (and others) are quite vigorously arguing that the “old” versions are “outdated” (as you claim), not everyone in the solar community agrees with them!
Remember, as we discuss in Section 2.2.1, the “sunspot number” is rather confusingly not “the number of sunspots”, but is actually an index which combines information on the number of sunspot groups, the number of individual sunspots and a factor based on who measured the sunspots. This means that compiling the dataset is surprisingly subjective. Indeed, this is why Clette et al. are arguing that several of the subjective decisions made for the other two versions are invalid. But, these arguments (while certainly plausible) are themselves somewhat subjective, and I know that several researchers disagree with some/all of their adjustments!
So, we think it is premature to conclude that the other datasets are “outdated”. This is part of the reason why we discuss all three versions in our review, not just “the latest”.
More importantly, we are not convinced that sunspot numbers are necessarily the best proxy for TSI. It’s true that they provide one of the longest datasets, and they certainly seem to be capturing an important aspect of solar variability… but they are a very indirect measure, particularly if the ACRIM group’s estimate of TSI is accurate.
Ronan Connolly October 4, 2015 at 12:58 pm
I haven’t seen many of the latter, in part because it’s very difficult to show a negative … which studies are you referring to?
And? …
I’m sorry, but whether they use high or low reconstructions is unimportant. The models are tuned, which makes their hindcasts meaningless.
Ronan, people have been looking for the effects of the ~11-year solar fluctuations for two centuries now … and there is still no clear, unassailable evidence that they do anything at all.
If you interpret that as being equality between the two propositions, I’m not sure what I can say.
Perhaps you care what the Tinkertoy models say or why they say it. I don’t.
TUNING! TUNING! They could just as easily be tuned to your dataset, but they weren’t.
Not sure what you mean by “at least to some extent significantly effected”. Either they are significantly affected, or they are not. What do you mean by “extent”? Temporal? Spatial? Conceptual?
Nor am I clear how you are measuring the effect of the UHI, other than by using only stations that by some criterion or another are “rural” … when you don’t even know whether they are “rural” or not, and you don’t have any clear definition of “rural” that applies to all the stations.
Yes, and it doesn’t agree with the satellite data, which has no significant UHI component … the part that you don’t seem to get is that simply because you call a station “rural”, that doesn’t mean that it doesn’t have problems, perhaps serious ones.
I dislike that automated homogenization algorithm, but then I’m not fond of doing anything automated to climate data.
It never was valid. It was always just the result of tuning. So your analysis doesn’t help there.
The models leave out or poorly parameterize the emergent phenomena which regulate the temperature to within a very narrow range (e.g. ± 0.3°C over the 20th century). So nothing they say has any more meaning than a monkey at a typewriter … and that means that all of your conclusions based on the fact that they are wrong are likely wrong …
I’m sorry, but I didn’t see that in your paper … and it is not clear what you are calling a “remarkable fit”. I also don’t know how “ACRIM” could update anything, since it’s a satellite dataset. Some links and examples would be very useful here.
Is it “plausible”? People since Hershel have thought so … but we still have no definitive data showing anything going on, despite two centuries of trying. If a failure to find any strong data in two hundred years makes something “plausible” on your planet, then it is indeed plausible.
Dunno … but I do know that I still haven’t heard anything about how the Bonferroni correction applies to the “remarkable fit” of one among twenty solar reconstructions …
I still don’t have any idea what you are calling the “remarkable fit”, how it was measured, or why you think it is “remarkable” …
Well, there’s a surprise.
First, while the compilation of the data was subjective, that does NOT mean that removing historical errors in the dataset are subjective. That makes no logical sense.
However, since you haven’t said what either you or the others think is wrong in the Clette et al. changes, I fear I can’t respond to those allegations.
I say it’s premature to say that the new reconstruction is wrong without letting us in on your specific objections to their changes, along with quotations of what they said that you think is wrong.
Regarding the “ACRIM group’s estimate”, all I find in your paper is this:
Scafetta is not a scientist, in that he has refused repeatedly to make his calculations public despite repeated requests. If Scafetta and Willson’s update is what you are calling the “ACRIM group’s estimate”, then the odds that it is unbiased, valuable, or accurate just went down stupendously … and that may be related to the fact that I don’t see a single citation to his work other than yours …
There’s a post of mine about Scafetta’s goofy claims here. In addition he was up to his ears in the infamous Copernicus-PRP affair. See Anthony’s post here, and don’t miss Anthony’s response to Nicola’s comment here. Depend on Scafetta at your own risk … and if his TSI reconstruction shows “correlations”, I’d be damn sure to examine his “correlations” with a microscope to see if they are real. Fortunately, not my problem, I don’t read anything of his these days. My rule is simple—no code, no data, no science.
In any case, the physical reasons that TSI is related to sunspots are well-known, and the historical sunspot records are available and detailed … so while no proxy is perfect, I’m not sure what you think might be a better proxy for TSI than sunspots.
Let me close by saying that I am greatly troubled by your lack of comment about the Bonferroni corrections, despite my explanation of why they are important. You keep saying things like that you considered ten or so different TSI reconstructions, and found one of them significant … well, duh, when you look at ten datasets, the odds of significance go through the roof … and you seem totally innocent of the consequence of that.
I am also distressed by the lack of any discussion of the effects of autocorrelation on the significance of the results. All of the datasets you are using are highly autocorrelated, and you appear to not know that this seriously changes your statistics.
Finally, I saw no discussion of the fact that an asymmetric cyclical dataset like sunspots or TSI will show totally bogus “significant correlations” with random red-noise data at an amazing rate. You need to do a Monte Carlo analysis to say anything at all about the statistical significance of results when you are using such a dataset … and the term “Monte Carlo” doesn’t appear anywhere in your paper.
All of this indicates to me that if I could say one thing to you, it is that you, like many others, are radically overestimating the significance of the things you are finding. From here, it appears that you desperately need a crash course in statistical corrections for autocorrelation, and in the Bonferroni correction, and in Monte Carlo analysis. You are getting badly fooled by what appear to be significant results, but which are nothing of the sort.
Best regards, and thanks for continuing the discussion.
w.
Hi Willis,
It’s quite late here and there’s quite a lot of points. I’ll try to address the main points, but apologies if my replies to individual sections end up a bit terse, or if I haven’t answered something on this round!
Willis Eschenbach October 4, 2015 at 10:45 pm
No, it was not a matter of what we “expected” them to do. Rather it was the case that the stations didn’t meet our initial ex-ante criteria. In our “Urbanization bias III” 2014 study, when we tried to identify those stations which met our a priori-defined selection criteria for that paper, i.e., fully rural stations with at least 95 years of data for the last 100 years, there were only 8 stations outside of the U.S. – 5 in Europe, 1 in Greenland, 1 in Canada and 1 in the Southern Hemisphere (New Zealand) – see Section 3.2 in that paper. Although it might be argued that the 5 European stations might possible be close enough for a regional estimate, we found a lack of consistency between all of the station records, and we concluded that – without access to the station histories for these stations it was difficult to determine their reliability (something which you apparently agreed with below).
We therefore recommended the collection of station histories for these (and other) stations should be carried out before using them for estimating regional temperature trends.
In January, 2015, we followed our own recommendation by personally collecting the station history for one of the 5 European stations, i.e., Valentia Observatory. Therefore, we were able to use that data for our new paper…
Willis Eschenbach October 4, 2015 at 11:02 pm
Where is your evidence that he “needed a new dataset to compare with”? He was using the standard datasets all the time, and didn’t seem to be running out of research ideas…
In the end, between our 4 regions, although we only used 10% of the Northern Hemisphere stations, we used nearly half of the fully rural stations in the Northern Hemisphere, and nearly 2/3 of the stations with data for the early 20th century.
http://s2.postimg.org/ohin5uujt/Fig_SI_03_station_use.jpg
These were the regions which had the greatest density of fully rural stations with relatively long records and/or useful station history/information. I agree that with careful work and/or
Here’s the comparison with UAH’s land only Northern Hemisphere data:
http://s22.postimg.org/j76hkiegx/Soon_vs_UAH_land_only_1880.jpg
http://s1.postimg.org/753bfc6e7/Soon_vs_UAH_land_only_1975.jpg
Unfortunately RSS don’t seem to do a land only estimate specifically for the Northern Hemisphere. However, as we discuss in Section 4.2, if we rescale our land-only estimate by 0.73 to reflect the dampened nature of ocean temperature trends, we can treat this as a crude “land+oceans” estimate. When we do this, we can compare our rescaled estimate to both RSS and UAH:
http://s17.postimg.org/3p546pynz/Soon_vs_satellites_land_oceans_1880.jpg
http://s11.postimg.org/t56rza7cj/Soon_vs_satellites_land_oceans_1975.jpg
I agree they’re not exact matches, particularly after about 2005, but to me they look fairly similar.
Willis Eschenbach October 5, 2015 at 12:10 am
There’s quite a lot of literature for both the former and the latter. Michael & I spent several weeks carefully reading & studying 100s from either side (Willie had already read most of them over the years, as that’s one of his main fields of expertise). In the paper, we tried to summarise the most representative ones, but if you have a look at the references section, you’ll see that’s still quite a lot!
Agreed! The main focus of our review paper was on “… the role of solar variability on Northern Hemisphere temperature trends since the 19th century”. So, we were mostly discussing possible influences on multidecadal time-scales, e.g., possible secular trends between solar cycles.
I’m also unimpressed by the current Global Climate Models. So, I don’t personally think much of their hindcasts either. But, apparently the IPCC do care! The results of the hindcasts were the main justification for their claim that the global temperature changes since the 1950s are mostly man-made.
We defined the “fully rural”/”intermediate”/”fully urban” categorisation in Section 3. We used the exact same categorisation as we did in our 2014 papers – see the last 2 paragraphs before Section 3.1.
See Section 3.1 for a discussion of urbanization bias in China. We discussed urbanization bias in the U.S. in considerable detail in our 2014 “Urbanization bias III” paper, so we just briefly discussed it in this new paper.
Note that it’s not so much “the effect of the UHI” that we’re interested in – it’s the effect of the expansion/growth of UHI at stations that have become urbanized.
See Section 5. We think the fits between the updated Hoyt & Schatten TSI reconstruction and our new mostly-rural Northern Hemisphere temperature reconstruction are ‘worthy of attention’. If the Hoyt & Schatten TSI reconstruction and our new temperature reconstruction are even partially accurate, then it implies long-term solar variability has played a substantial role in temperature trends since the 19th century.
However, see also Section 1.1 and Section 2, where we stress that “correlation doesn’t imply causation” and that there are a lot of uncertainties about the TSI reconstructions which have not been satisfactorily resolved!!! This is why we repeatedly tell the reader to treat this result with considerable caution.
Personally, I think their proposed changes are quite plausible and their rationale for the changes seems quite reasonable. Indeed, we noted that their adjustments seemed to slightly reduce the apparent differences between sunspot numbers and the solar cycle length proxies. But, as I explained yesterday, in our review, we were trying to accurately summarise the various perspectives among the community, and avoid forcing our own personal opinions down the reader’s throat!
We discuss the debate between the three satellite TSI groups (ACRIM, PMOD and RMIB) in Section 2.1. It’s worth reading carefully – in my opinion, it’s quite representative of the whole debate. In particular, have a look at the quotes at the end of the section, e.g.,
vs.
Clearly, both groups have very strong views on the relative importance of greenhouse gas emissions vs. solar.
All three groups base their TSI composites predominantly on the ACRIM satellites. Willson is the P.I. in charged of these satellites.
The ACRIM group use the data mostly unadjusted. Their TSI estimate implies TSI increased between cycles from 1980-2000 and then decreased. The PMOD group apply several “corrections” to the ACRIM data first. Their TSI estimate implies a fairly continuous decrease in TSI between cycles since 1978.
To me it is unclear which (if any) of the satellite estimates is most reliable.
The ACRIM group also recently updated the Hoyt & Schatten, 1993 TSI reconstruction (http://www.atmosp.physics.utoronto.ca/people/guido/PHY2502/articles/solar-activity/Hoyt-Schatten.pdf) using the ACRIM composite for the satellite era component instead of the NIMBUS dataset which Hoyt & Schatten had used.
Hoyt was the P.I. for the NIMBUS satellite, which was the first satellite to systematically record TSI – it started 2 years before the first ACRIM satellite.
Scafetta joined the ACRIM group a few years back. I also found several of his studies (particularly the ones on his theories about barycenters) to be problematic. A lot of them seemed to me to be little more than curve-fitting.
If you read Section 2, one of our major conclusions is that there are still many unknowns, and we still don’t know which (if any) of the TSI reconstructions are most reliable. So, I totally agree with you there. Yes, researchers who are simply picking the reconstruction which gives them “the best fit” are artificially inflating their P values, and in such cases, as you say, they should apply the Bonferroni correction, for instance. Simmons et al., 2011 (http://pss.sagepub.com/content/22/11/1359) provide a great example of a similar phenomenon in the field of psychology.
Did you read Section 1.1?
It’s probably important to distinguish between the terms “statistically significant”, “significant correlations” and the more general term “significant”. The use of the first two implies you have carried out some statistical tests on your result. The latter term can be used in a more casual manner, e.g., when you are merely stating something is not insignificant!
Could you clarify exactly which “things you are finding” you are referring to? The solar variability aspect of this paper was mostly a review of the literature. It’s true that in Section 5.1, we did carry out some statistical tests of the Hoyt & Schatten
P.S. My PhD was on Monte Carlo simulations (albeit in a different field – polymer physics
Ronan, let me offer an illustration of how wrong standard statistical analysis is in the world of climate. Here is the sunspot data and the HadCRUT global temperature data. We want to determine if there is a relationship.

Here is the bog-standard analysis of the linear relationship of temperature as a function of sunspots.
Coefficients: Estimate Std. Error t value P-value (Intercept) -0.1436108 0.0109595 -13.104 < 2e-16 *** Sunspots 0.0004062 0.0001002 4.055 5.21e-05 *** --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ' 1 Residual standard error: 0.304 on 1975 degrees of freedom Multiple R-squared: 0.008257, Adjusted R-squared: 0.007755 F-statistic: 16.44 on 1 and 1975 DF, p-value: 5.208e-05Now, according to that result, the relationship between temperature and sunspots is indeed very very significant, a p-value of 5E-5, so the case is clearly proven that it’s the sun … but is it really?
There are a couple of ways to determine that. You can use the method of Quenouille to adjust your significance calculations.
According to that method, the p-value is 0.11, far from significant at the usual significance level of p<0.05.
Or you can do a Monte Carlo analysis by generating similar “pseudo-data" which is random, and seeing how well correlated the pseudo-data is with the sunspot data. The sunspots and HadCRUT data have a correlation of 0.09, which the standard analysis says is significant, but Quenouille says no. Here’s a sample of my pseudodata, along with the actual HadCRUT data:

So what do we get when we look at the correlation of the sunspots with this pseudo-data? Here’re the results for 1,000 instances.

Recall that the correlation of the sunspot data with the actual HadCRUT data is 0.09 … and as you can see, that is not unusual in the slightest.
Finally, I do what I call a “straight line comparison”. To do that, we simply compare the correlation of sunspots with the HadCRUT temperature (cor=0.091) on the one hand, to the correlation of sunspots with a straight line on the other hand … and that straight line correlation, my friend, is 0.098. The correlation of sunspots is BETTER with a straight line than with the HadCRUT data.
All of which demonstrates why I said that asymmetrical cyclic data like sunspots give bogus correlations with about anything … and why I said above that standard stats are not just wrong but way, way wrong in this case.
Despite the fact that the standard cookbook math says that the relationship between sunspots and HadCRUT temperature is highly significant at a p-value of 0.00005, Quenouille and the Monte Carlo analysis and the straight line comparison clearly demonstrate that no way, it’s not significant at all …
Now, with all of that in mind, let me suggest that you re-examine Scafetta’s claim of an “apparently strong correlation" of TSI with something or other … my guess is that you will find an apparent correlation, but not a strong correlation in any sense.
w.
Ronan, thanks for your detailed reply. I’ll answer it in pieces. First, I’d said:
You replied:
Ronan, you claimed you didn’t use them because inter alia they “implied slightly different trends”. Therefore, you are looking at one of your proposed new reconstruction’s OUTCOMES (the trend of the the reconstruction) to determine which stations to include.
And that is indeed data snooping. You are looking, not at the metadata for the station (e.g. rural, length of record, etc.) to choose your stations. You are looking at the trend of the station data in order to decide whether to use it … sorry, but you can’t do that.
Here’s an example where it might be more obvious. Suppose I want to show that the earth has really warmed more than people say. So I look at the trends of the stations, JUST AS YOU DID, and I decide not to use some of them because of their trends, JUST AS YOU DID.
And guess what … I am able to prove what I claimed!
But only because I snooped the data, and didn’t restrict myself to looking only at the metadata.
Do you see the problem now? If not I can give other examples.
w.
Ronan Connolly October 5, 2015 at 4:26 pm
Same objection, which you have not addressed. If you only used half the fully rural stations, why didn’t you use the rest? I simply don’t believe that every single suitable station just happened to be in a couple of countries (and you’ve already said your arctic stations were unsuitable but you used them anyway).
w.