Guest essay by Johannes Herbst
There is a much discussed graph in the blogosphere from ‘Tamino’ (Grant Foster), which aims to prove that there is no delay or pause or decline in global warming.
He states: Twelve of sixteen were hotter than expected even according to the still-warming prediction, and all sixteen were above the no-warming prediction:
Let’s get a larger picture:
- We see the red HADCRUT4 graph, coming downwards a bit from 1960 to 1975, and inclining steeper beyond 2000, with a slight drop of about the last 10 years.
- We see a blue trend, rising at the alarming rate of 0.4°C within only one decade! This was the time when some scientists started to worry about global warming.
- We see the green trend, used by the blogger Tamino in the first graphic, rising less than 0.1°C per decade.
- Below we see the Sunspot Numbers, pulsing in a frequency of about 11 years. Comparing it with the red temperature graph, we see the same pattern of 11 years pulsing. It shows clear evidence that temperature is linked to the sunspot activity.
Tamino started his trend at high sun activity and it stopped at low activity. Therefore the weak increase during 18 years.
Which leads us to the question: How long should a time be for observing climate change? If we look at the sunspot activity and the clear pattern it produces in the temperature graph, the answer is: 11 years or a multiple of it.
Or we can measure from any point of:
·high sun activity to one of the following
·low sun activity to one of the following
·rising sun activity to one of the following
·declining sun activity to one of the following
to eliminate the pattern of sunspot numbers.
Let’s try it out:
The last point of observation of the trend is between 2003 and 2014, about 2008. But even here we can see the trend has changed.
We do not know about the future. An downward trend seems possible, but a sharp rise is predicted from some others, which would destroy our musings so far.
Just being curious: How would the graph look with satellite data? Let’s check RSS.
Really interesting. The top of both graph appears to be at 2003 or 2004. HADCRUT4 shows a 0.05°C decline, RSS a 0.1°C per decade.
A simple way for smoothing a curve
There is a more simple way for averaging patterns (like the influence of sunspots). I added a 132 months average (11 years). This means at every spot of the graph all neighboring data (5.5 years to the left and 5.5 years to the right) are averaged. This also means that the graph will stop 5.5 years from the beginning or the end. And voila, the curve is the same as with our method in the previous post to measure at the same slope of a pattern.
As I said before the top of the curve is about 2003, and our last point of observation of a 11 years pattern is 2008. From 2008 to 2003 is only 5 years. This downtrend, even averaged, is somehow too short for a long time forecast. But anyway, the sharp acceleration of the the 1975-2000 period has stopped and the warming even halted – for the moment.
Note: I gave the running average graph (pale lilac) an offset of 0.2°C to get it out of the mess of all the trend lines.
If Tamino would have smoothed the 11years sun influence of the temperature graph before plotting the trend like done here at WFT, his green trend would be would be the same incline like the blue 33 year trend:
Even smoother
Having learned how to double and triple smooth a curve, I tried it as well on this graph:
We learned from Judith Curry’s Blog that on the top of a single smoothed curve a trough appears. So the dent at 2004 seems to be the center of the 132 month’s smoothed wave. I double smoothed the curve and reached 2004 as well, now eliminating the dent.
Note: Each smoothing cuts away the end of the graph by half of the smoothing span. So with every smoothing the curve gets shorter. But even the not visible data are already included in the visible curve.
According to the data, after removing all the “noise” (especially the 11 year’s sun activity cycle) 2004 was the very top of the 60 years sine wave and we are progressing downwards now for 10 years.
If you are not aware about the 60 years cycle, I just have used HADCRUT4 and smoothed the 11 years sunspot activity, which influences the temperature in a significant way.
We can clearly see the tops and bottoms of the wave at about 1880, 1910, 1940, 1970, and 2000. If this pattern repeats, the we will have 20 more years going down – more or less steep. About ten years of the 30 year down slope are already gone.
One more pattern
There is also a double bump visible at the downward slopes of about 10/10 years up and down. By looking closer you will see a hunch of it even at the upward slope. If we are now at the beginning of the downward slope – which could last 30 years – we could experience these bumps as well.
Going back further
Unfortunately we have no global temperature records before 1850. But we have one from a single station in Germany. The Hohenpeissenberg in Bavaria, not influenced from ocean winds or towns.
http://commons.wikimedia.org/wiki/File:Temperaturreihe_Hoher_Pei%C3%9Fenberg.PNG
Sure, it’s only one single station, but the measurements were continuously with no pause, and we can get somehow an idea by looking at the whole picture. Not in terms of 100% perfection, but just seeing the trends. The global climate surely had it’s influence here as well.
What we see is a short upward trend of about ten years, a downward slope of 100 years of about 1°C, an upward trend for another 100 years, and about 10 years going slightly down. Looks like an about 200 years wave. We can’t see far at both sides of the curve, but if this Pattern is repeating, this would only mean: We are now on the downward slope. Possibly for the next hundred years, if there is nothing additional at work.
The article of Greg Goodman about mean smoothers can be read here:
Data corruption by running mean ‘smoothers’
==================================
Johannes Herbst writes at: http://klimawandler.blogspot.de/
Like RichardLH, I have also been banned by Tamino so I’ m certainly not of his followers. However, it needs to be said that Tamino’s graph is not a trick. There are effectively 2 ways of interpreting the temperature data over the past 30-odd years, i.e.
1. There has been NO significant warming since 1997 (or 1998 or whatever)
2. Warming of ~0.17 degrees per decade has continued since 1997 (or 1998 or whatever)
Depending on the question asked (or the hypothesis proposed) BOTH statements can be TRUE. To illustrate consider the dataset which shows least warming, i.e. RSS. Since 1997 the RSS trend is -0.013 ±0.201 °C/decade (2σ)
If our NULL Hypothesis is that there has been no warming since 1997 then we would clearly have to accept that as the ZERO trend lies within the 2-sigma confidence interval. Fine – but what if the NULL hypothesis was that warming was continuing at the rate of 0.17 degrees per decade. Again there is no reason to reject the NULL hypothesis since a trend of 0.17 degrees per decade also lies within the 2-sigma CI.
It will be some time before we see if there is a genuine change in the trend. For what it’s worth, I doubt very much there will be a strong cooling trend but the warming trend is likely to be less than that projected by IPCC models.
… and you’d need a variable lag as one of the parameters at least.
Then there’s the added problem that the response, once isolated, may not be simple linear fn of the forcing even with a lag. It will more likely be a combination of the forcing and its derivative in a way that corresponds to a laplacian decaying exponential response (ie a linear relaxation).
http://climategrog.wordpress.com/?attachment_id=399
Even that is assuming that there is one single time constant in the climate system response to a particular forcing.
This is really NON trivial stuff. That is why the various attempts at linearly regressing various “forcings”, even multivariate linear regressions, get poor results.
Then naive authors ( such as the Great “Professor” Tamino ) start thinking they have derived the proportion of all the major drivers and what’s left must be AGW.
With all the possible drivers, possible response function plus possibly internal oscillations, you need some many free variables you will always manage to get a fit where the residual can credibly be attributed to the sampling error.
Really all this kind of exercise can do is suggest some possible relationships and interactions that can then be used as a starting point for further study of mechanisms etc.
It would be really neat if climate was determined to about 90% by one major driver that could be estimated by simple linear regression of a few hypothetical response functions. Sadly it’s a bit more complicated.
pyromancer76 says:
February 7, 2014 at 7:16 am
we seem to be in some kind of solar minimum (again), but what was the sun “doing” during the 100-year interim?
It was doing the same thing as it did in the 19th century and in the 18th: going from a minimum to a maximum and then back to the minimum again. Did climate do the same? I think not.
greg says:
February 7, 2014 at 1:26 pm
neither is one “shorthand’ for the other.
If I use one as a shorthand for the other, then it is a shorthand for the other in my usage. TSI is usually taken as an indicator of the energy received, so you were nitpicking. You want to nitpick some more?
If TSI (power) in some way affects mean surface temperature it is not going to correlate with temperature directly unless the whole effect equilibrates in a time-scale much shorter than a solar cycle.
Tell that to the author of the post who claimed that every temperature ‘pulse’ was caused by a similar sunspot pulse. Now solar activity has for three centuries been following the same course: (minimum, maximum, minimum), and if you claim the time for equilibrium is much longer than a sunspot cycle, then you cannot correlate directly with the sunspot cycle as was done. In any case, your response was just handwaving, no numbers, no mechanism, etc.
Willis Eschenbach says:
February 7, 2014 at 1:01 pm
Leif, while I generally agree with you, this claim (that in the long run TSI rules the temperature) ignores the obvious—the thermally driven response of cloud increase, which cuts down the amount of energy hitting the earth. TSI goes up … clouds go up … temperature stays the same.
I did say just after my comment, that my claim would hold unless there were a sort of thermostat [which might well be, but is not generally accepted by everybody].
Russ R. says:
February 7, 2014 at 1:45 pm
For three solar cycles the oceans release sub-surface heat into the LT, which we record. For three cycles additional heat is transported below the surface, and cold water is brought to the surface, where it is measured as surface temp. I don’t have a reason why, but it is quite clear to me, that there is a link, between the sun, the oceans, and the temps of the LT.
I have my doubts about solar cycles being relevant but I agree that the oceans are the primary driver through upwelling of cold water. The Meridional Overturning Circulation (MOC) is a current that circles the globe. When it speeds up we get more of the warm water on the surface replaced with cold upwelling water. When it slows down the warm surface water stays on top longer. The atmosphere is always seeking equilibration with the oceans. I suspect this is the driver of both the ~60 year cycle and the ~900 year cycle. However, I also suspect the causes of the two cycles may be different.
“”He{Grant Foster} states : Twelve of sixteen were hotter than expected even according to the still-warming prediction, and all sixteen were above the no-warming prediction:””
——————————————————————————————————
It is of interest to note how sure he is that 12 of 16 were hotter than ‘expected’. I assume this means that Foster presumes to know the average temperature of any given future year using some formula. Does he have his list of predictions for the next dozen years that we could see, so that we would be able to ascertain how accurate his initial predictions were? How far above the anticipated predictions were the actual temperatures?
Cross-correlation … never even makes it to a correlation of 0.10 … using the same data from WoodForTrees as in the head post.
w.
Gareth Phillips says:
February 7, 2014 at 3:12 am
If so, the conveyor belt of storms experienced since last year in the UK will have to be accepted as quite normal as well as other climatic changes yet to be seen.
———————————————————————————-
Excuse me for being impolite, but b.s..
Greg Goodman says:
February 7, 2014 at 3:58 am
————————————-
That was helpful regarding the Bavarian temp record.
The Bavaria record shows a strong resemblance between the last 20 years 1990/2010, and the period 1790/1810. I have noticed similar before on different long term graphs of historical temps.
Straight line linear regression is just plain ignorance. When we already know that climate processes rise and fall, putting straight lines on any rise-and-fall system, chaotic or not, gives absolutely nothing usable for the long term. Between now and 2100 there will be several ups and downs. And we could play with straight lines till we turn blue in the face, and it will tell is NOTHING that won’t be wrong within 5-15 years.
John Finn says:
February 7, 2014 at 4:51 pm
“Like RichardLH, I have also been banned by Tamino so I’ m certainly not of his followers.”
Welcome to the club (or am I joining you?) 🙂
“Since 1997 the RSS trend is -0.013 ±0.201 °C/decade (2σ)….It will be some time before we see if there is a genuine change in the trend.”
Using Linear Trends (a discrete function) is always going to be a poor second cousin to using a true continuous function such as a filter if you are trying to predict what a data set will do in the near future.
Using LOWESS (a continuous OLS function built on short LTs) is going to be less useful than an S-G filter (using OLS to an underlying curve). IMHO.
All estimation devices suffer from the same problem however. They are only estimations. They need backing with some invariant core backbone such as I try to do with the CRTM filters I have applied to the data. This at least allows a reasoning for the estimation parameters then chosen.
http://i29.photobucket.com/albums/c274/richardlinsleyhood/HadCrut4Monthly11575Lowpass1575SGExtensions_zps48569a45.gif
Steve Garcia says:
February 8, 2014 at 12:51 am
“Straight line linear regression is just plain ignorance.”
‘Linear Trend’ = ‘Tangent to the curve’ = ‘Flat Earth’ 🙂
Willis Eschenbach says:
Cross-correlation … never even makes it to a correlation of 0.10 … using the same data from WoodForTrees as in the head post.
===
That would be a better place to start but like I said in some detail above, any single variable analysis will not show a relationship even it if exists unless it really the dominant signal. Since many appear to think it is you go someway to disproving that misconception.
FWIW , here is the spectrum of HadCruft4 that I did some time ago.
http://climategrog.wordpress.com/?attachment_id=121
Main peak is lunar not SSN,
The other thing here is that HadCruft4 is really not the right place to start looking anyway. It is a composite dataset of SST and land temps ( E&OE ) . Land is twice as sensitive to radiative change as ocean and has a strong N/S bias. Land will also have a notably different time constant in its response to radiative change.
Since surface temp is some clumsy proxy for heat content which ignores many pertinent variables, this sort mixed data will be further muddy the waters.
Most of reason that climate science makes no progress on attribution is that their methods and data would not find one even if is were present.
In view of the prevalent mode of thought at UEA and other guardians of the data, this may well be intentional.
I don’t think the solar component to surface temps is strong but the plot you provide is not the right way to go looking for one. SST only would be a better start.
From the volcano stack plots I did I would conclude that tropics are fairly immune to radiative changes and that any solar signal will be more likely to show in extra-tropical SST.
However, the mains common periodicity in SST seems to be the lunar circa 9y not solar.
http://climategrog.wordpress.com/?attachment_id=56
Greg Goodman says:
February 8, 2014 at 2:04 am
“However, the mains common periodicity in SST seems to be the lunar circa 9y not solar.”
I do wonder if the ~60 year signal has a Lunar component/source as well. I don’t have the data or skills to find it though.
taking 9.15 as a typical value for the “lunar” component that seems very widespread, and 64 as the long cycle leads to 10.68 as a frequency that could lead to an interference “beat” frequency. (That is amplitude envelop like what we hear with musical or acoustic “beats” , rather that the modulation frequency which is half as fast).
10.68 does not actually happen ( and 60 is only approximate also) but could be an averaged effect of solar cycle lengths that tend to bunch around 10.4 and something over 11. Obviously a circa 60 year period covers about six cycles and none of this is nice pure cosines.
The beat frequency is simply the difference so in terms of periods you need to look at:
1/p = 1/p1-1/p2
That gives a means to derive the 60 year figure, you then need to explain what the ‘lunar’ bit is physically and why the it is the envelop of the beats that affect climate instead of the fast cycles just averaging out to zero.
As a quick stab at mechanisms I’d say lunar tidal displacement of water mass , hence thermal energy in and out of tropics, ie longer period tides. Willis’ tropical governor would restore SST in tropics so there will be net changes in heat input to Earth system globally.
Tropics are probably fairly immune to solar changes but ex-tropics less so. The two will add to create an interference pattern.
The non linear response in the tropics means it will not just average out over a full cycle. Hence the amplitude of oscillations will probably determine the long term net effect,
Now that’s just a sketch of how such a period _could_ come about but I would not claim any more than that.
Greg:
“That gives a means to derive the 60 year figure, you then need to explain what the ‘lunar’ bit is physically and why the it is the envelop of the beats that affect climate instead of the fast cycles just averaging out to zero. ”
I suspect that the ‘beat’ is with the Solar, exactly as it is with all tides. I am thinking that the paths that the two main tidal agents follow here on Earth, both directly overhead and at ~60 degrees to that path, are the patterns to look for.
Why ~60 degrees? Because I think that this is a possibly important factor – http://i29.photobucket.com/albums/c274/richardlinsleyhood/Tidalvectors_zps4fd5800f.png
The problem is that I cannot find a data set online that will allow those paths and combinations of those paths to be constructed/downloaded.
Sorry, make that 45 degrees – I’ve got 60 on my brain 🙂
A little signal processing theory and possible application to this problem: If we take the “channel” as the 60 year cycle present in nature, then the Nyquist rate – how many conclusions we are allowed to draw, if you will, or in control theory, how often we’re going to provide some feedback to control the climate – is about 30 years apart (in other words 2x the bandwidth of the channel, the theoretical limit). From practical experience with sampling oscilloscopes, you generally want 4x the Nyquist criterion to accurately reproduce a signal, so that means we should draw conclusions about every 120 years. Since we have accurate satellite data since 1979, we have to wait until 2099 to determine if there’s really a global warming signal or whether we were just drawing bad conclusions from not seeing the whole cycle a couple of times.. I’m leaving aside the problem that there might be an overlay of a 200 year cycle on top of the 60 year cycle of course – and probably many other cycles as well – though in 120 years we should accurately perceive any cycle that’s less than 60 years long, no matter how many are overlayed on the signal.
rgbatduke says: February 7, 2014 at 12:39 pm
Thank you Sir for your appeal for courtesy in debate.
A few observations:
I do not agree with Gareth but have gained some new knowledge from his conversations with you, Richard(s), etc.
While I do not necessarily agree with Johannes’ conclusions he has provoked thought and I cannot dismiss everything he says simply because he is apparently not an expert statistician.
I try to maintain an open mind on this subject because climate science is in its infancy and we really do not even fully understand what drives what. The common consensus that “CO2 drives temperature” is contradicted by the observation that CO2 lags temperature at all measured time scales. I suggest that the future cannot cause the past.
Many of the comments here seem to assume that the Hadcrut4 temperature record is accurate. My work circa 2002 suggested a probable warming bias in Hadcrut3 of about 0.07C per decade since 1979. How far back in time this applies is a matter of conjecture. Is Hadcrut4 significantly better than Hadcrut3? Probably not.
The modern thermometric global Surface Temperature records are not sufficiently accurate for our scientific discussion, in my opinion.
I suggest that the 1930’s were as warm or warmer than today in the continental USA, and perhaps even globally as well. Earth was certainly warmer than today during the Medieval Warm Period and earlier warm periods. There is nothing unusual about today’s climate or weather, and there is no discernable humanmade influence to date.
I have maintained the same position on the alleged global warming crisis since about 1985, which we formally stated in 2002 as follows:
“Climate science does not support the theory of catastrophic human-made global warming – the alleged warming crisis does not exist.”
http://www.apegga.org/Members/Publications/peggs/WEB11_02/kyoto_pt.htm
Regards to all, Allan
This is a no-brainer. Download the data from HADCRUT4 and run slopes to earlier and earlier dates back from the latest entry. What you will find is that the earliest date that still can be found that delivers a zero or negative slope is about November of 2000. (I’m doing this from memory – I think I posted this same stuff here about a week ago – I’m away from home on a hotel machine) but prior to that November date all the slopes will be positive. Doesn’t matter what B.S. Grant Wood – Taminio or the Easter Rabbit come up with.
Greg Goodman says:
February 8, 2014 at 2:04 am
Thanks for the reply, Greg … unfortunately, all I can say is huh? That’s not true in any sense. As one of many examples, if we investigate the tide for one of the minor components, that component will stand out even though there are other more dominant signals in the mix.
How do I know that? Well, because I’ve done it. Among other examples, when I lived in the South Pacific, I ran a remote shipyard, and I had to generate my own tide tables because there were none available. It’s an interesting exercise, Greg, you should give it a try.
In any case, your claim would be greatly strengthened by an actual example. I’ve given you an example, the tides, where we can “show a relationship” of a minor signal … give us a natural observational dataset that contains a major and a minor signal, and give us the analysis that you say can’t show the minor signal. Then we’ll have something to discuss.
Let me go back to where this started. In the head post Johannes said:
I called BS on that statement. I said that was not true. I’ve shown a variety of evidence in support of that, showing no correlation between temperature and sunspots, using the exact data he used for the head post.
On the other hand, neither you nor anyone else has put forward a scrap of evidence to support Johannes’s statement. You saying “any single variable analysis will not show a relationship even it if exists” is not evidence in support of his statement.
w.
Greg Goodman says:
February 8, 2014 at 2:04 am
That’s useless. Your graph has not once scrap of explanatory text with it. Go take a look at it, I’ve left a comment.
I hate following a link to find it’s a tease, with no data, no code, no explanation, no underpinnings. Please don’t do it. For heavens sake, you didn’t even put in the damn units!
w.
Greg Goodman says:
February 8, 2014 at 2:04 am
Perhaps you are right. However, since HadCRUT is 70% SST, if there were a sunspot signal in SST we’d see it in HadCRUT. We don’t.
It’s an interesting analysis, Greg. Again, without a link to code and data it’s just advertising … but it’s interesting advertising.
I have no problem with the nine-year cycle, because it’s present in the tides and those forces are well understood. In the Keeling and Whorf paper I’ve discussed recently, the authors say:
w.
Greg Goodman says:
February 8, 2014 at 2:04 am
FWIW , here is the spectrum of HadCruft4 that I did some time ago.
http://climategrog.wordpress.com/?attachment_id=121
Main peak is lunar not SSN,
As the period of lunar nutation is 18.6 years [half of it is 9.3] how can you claim the peak is lunar? So your ‘FWIW’ is ‘not much’.
Willis: “You saying “any single variable analysis will not show a relationship even it if exists” is not evidence in support of his statement.”
Neither did I claim it indicated support for his statement. On the contrary I have posted quite a bit of comment saying how he not looking a broad enough sample and even over three cycle you can see the phase shift and … well read it.
lsvalgaard says:
As the period of lunar nutation is 18.6 years [half of it is 9.3] how can you claim the peak is lunar? So your ‘FWIW’ is ‘not much’.
====
8.85 interference with 18.6/2 gives 9.06 . The interference pattern produces 9.06 modulated by about 356 years. So what comes out on on spectral analysis is the circa 9.06.
I think this is the same signal as Scafetta found in JPL Horizon data and showed it was due to the presence of the moon by comaparing spectra of motion of E-M barycentre. BEST published a similar frequency for land ST last year.
Cross-correlation of N. Altantic and extra tropical N.Pacific SST (ie not PDO) gives very close to the same thing:
http://climategrog.wordpress.com/?attachment_id=755
Yes, it merited a bit more explanation but I get tired of repeating it each time.
Hope that explains the lunar attribution.