Chylek Imitates Ouroboros

Guest Post by Willis Eschenbach

Bob Tisdale has a detailed post on the new 2014 paper entitled “The Atlantic Multidecadal Oscillation as a dominant factor of oceanic influence on climate” by Chylek et al. Nic Lewis also did a good analysis of the paper, see the Notes below for the links. I have a different take on it than theirs, one which centers on the opening statement from their abstract:

ABSTRACT: A multiple linear regression analysis of global annual mean near-surface air temperature (1900–2012) using the known radiative forcing and the El Niño–Southern Oscillation index as explanatory variables account for 89% of the observed temperature variance. When the Atlantic Multidecadal Oscillation (AMO) index is added to the set of explanatory variables, the fraction of accounted for temperature variance increases to 94%. …

They seem impressed with a couple of things. The first is that their four aggregated forcings of greenhouse gases (GHGs), aerosols, volcanic forcings, and solar variations, plus an ENSO dataset, can emulate the global average temperatures with an adjusted R^2 of 0.89 or so. The second thing that impresses them is that when you add in the AMO as an explanatory variable, the R^2 jumps up to 0.94 or so … I’m not impressed by either one, for reasons which will become clear.

forcings used in chylek et al 2014Figure 1. Forcings used in the Chylek et al. analysis of the Atlantic Multidecadal Oscillation. Note the different scales in each panel.

There are several problems with the analysis done in Chylek 2014. Let me take the issues in no particular order.

PROBLEM THE FIRST

Does anyone but me see the huge issue inherent in including the Atlantic Multidecadal Oscillation (AMO) Index among the explanatory variables when trying to emulate the global surface temperature?

Perhaps it will help if I post up the explanation of just how the AMO Index is calculated …

From their link to the AMO dataset (see below)…

The [AMO] timeseries are calculated from the Kaplan SST dataset which is updated monthly. It is basically an index of the N Atlantic temperatures. …

Method:

Use the Kaplan SST dataset (5×5).

Compute the area weighted average over the N Atlantic, basically 0 to 70N.

Detrend that time series

Optionally smooth it with a 121 month smoother.

In other words … the AMO is just the temperature of the North Atlantic with the trend removed.

So let me ask again … if we’re trying to emulate the “global annual mean near-surface air temperature for the period 1900-2011″, will it help us if we know the detrended North Atlantic temperature for the period 1900-2011 … or is that just cheating?

Me, I say it’s cheating. The dependent variable that we are trying to emulate is the global surface temperature. But they have included the North Atlantic temperature, which is a large part of the very thing that they are trying to explain, as an explanatory variable.

But wait, it gets worse. The El Nino index that they use is a fairly obscure one, the “Cold Tongue Index”. It is described as follows (emphasis mine):

The cold tongue index (CTI) is the average SST anomaly over 6N-6S, 180-90W (the dotted region in the map) minus the global mean SST.

There are a number of El Nino indices. One group of them are the detrended average of the sea surface temperatures in various areas—El Nino 1 through El Nino 4, El Nino 3.4, and the like. There is also the MEI, the Multivariate ENSO Index. Then there are pressure-based indices like ENSO, based on the difference in pressure between Tahiti and Darwin, Australia.

There’s an odd wrinkle in the cold tongue index (CTI), however. This is that the CTI is not detrended. Instead, they subtract the global average sea surface temperature (SST) from the average temperature in the CTI area of 6°N/S, 180° to 90° W.

But this means that they’ve included, not just the average temperature of the CTI area, but also the entire global SST as a part of their explanatory variable, because:

CTI Index = CTI Sea Surface Temperature – Global Mean Sea Surface Temperature

I ask again … if you are trying to emulate the “global annual mean near-surface air temperature for the period 1900-2011″, will it help if an explanatory variable contains the global mean sea surface temperature for the period 1900-2011 … or again, is that just cheating?

I have to say the same as I said before … cheating. Using some portion of this year’s global temperature data (e.g. North Atlantic SSTs or CTI SSTs or global SSTs) to predict this year’s global temperature data is not a valid procedure. I’m sure my beloved and most erudite friend Lord Monckton could tell us the Latin name of this particular logical error, but Latin or not … you can’t do ‘dat …

Which is why, although the authors seem to be impressed that including the AMO increased the adjusted R^2 up to 0.94, I’m not impressed in the slightest. You can’t use any part of what you are trying to predict as a predictor. See how the AMO index (bottom right, Fig. 1) goes down until 1910, then up until 1940, down until 1970, and then up again? Those are the North Atlantic version of the very swings in temperature that we are trying to explain, so you absolutely can’t use them as an explanatory variable.

PROBLEM THE SECOND

Let’s look at just the forcings used in the climate models, setting aside the ENSO and AMO variables. Chylek 2014 uses the GISS forcings, which are composed of the following separate datasets:

GISS forcings used in chylekFigure 1a. The ten categories of forcing in the GISS forcing dataset. Note the different scales for each panel.

Now, for anybody that thinks that e.g. ozone levels in the atmosphere actually look like that … well, seems highly doubtful. But while that is a problem in and of itself, that’s not the problem in this context. The problem here is that all of these are measured in watts per square metre (W/m2). As a result they should all have the same effect … but Chylek et al. do a strange thing. They add together the well-mixed ghgs plus ozone plus stratospheric H2O into one group they call “GHGs”. Then they put reflective aerosols, aerosol indirect, black carbon, and snow albedo into a second group they call “Aerosols”. Volcanic forcing are treated as a third separate group, solar is the fourth, and land use is ignored entirely. This grouping is shown in Figure 1 above.

Then each of these four groups (GHSs, Aerosols, Volcanoes, and Solar) gets its own individual parameter in their equation … but this means that a watt per square metre (W/m2) from aerosols and a W/m2 from solar and a W/m2 from GHGs all have a very, very different effect … they make no effort to explain or justify this curious procedure.

PROBLEM THE THIRD

Here’s an odd fact for you. They are impressed that they can get an R^2 of 0.88 or something like that (if they cheat and include the entire global SST within the “explanatory” variables of their model). I can get close to that, 0.87. However, let’s start by calculating the R^2 of a much simpler model … the linear model. Figure 2 shows the GISS Land-Ocean Temperature Index (LOTI), and a straight-line emulation. The odd fact is the size of the R^2 of such a simplistic model …

chylek emulations Linear_TrendFigure 2. The simplest possible straight-line model. Black is GISS LOTI, red is the emulation.

Note that the R^2 of a straight line is quite high, 0.81. So their correlation of 0.88 … well, not all that impressive.

In any case, here are a few more emulations, with their corresponding adjusted R^2. First, Figure 3 shows their group called “aerosols” (AER) along with the volcanic forcing (VOL):

chylek emulations AER VOLCFigure 3. Emulation using the Chylek groups “Aerosols” (AER) and “Volcanic Forcing” (VOL). Note that watt for watt, the aerosols have about six times the effect of the volcanoes.

Now, even this bozo-simple (and assuredly incorrect) emulation has an adjusted R^2 of 0.854 … or, if you don’t like the use of aerosols, Figure 4 shows the same thing as Figure 3, but with GHGs in place of aerosols:

chylek emulations GHG VOLCFigure 4. Emulation using solely GHGs (GHG) and volcanoes (VOLC). Note that watt for watt, the GHGs have about three times the effect of the volcanoes.

There are a couple of issues revealed by this pair of analyses, using either GHGs or aerosols. One is that you can hardly see the difference between the two red lines in Figures 3 and 4. Obviously, this means that getting a good-looking match and a fairly impressive-sounding adjusted R^2 means absolutely nothing about the underlying reality.

Another issue is the difference between the strengths of the supposedly equivalent W/m2 values from GHGs, aerosols, and volcanoes.

Having seen that, let’s see what happens when we use all of the Chylek forcings except the cheating forcings (ENSO and AMO). Figure 5 shows the emulation using the sun, the aerosols, the volcanoes, and the greenhouse gases:

chylek emulations GHG AER VOLC SOLFigure 5. Emulation using the four Chylek at al. groupings (greenhouse gases GHG, aerosols AER, volcanic VOLC, and solar SOL) of the ten GISS forcings.

Note that again, watt for watt the volcanoes are only about a third of the strength of the GHGs. The solar forcings are quite strong, presumably because the solar variations are quite small … which highlights another problem with this type of analysis.

So that’s the third problem. They are giving different strengths to different types of forcings, without any justification for the procedure. Not only that, but the variation in the strengths is three to one or more … I see no physical reason for their whole method.

PROBLEM THE FOURTH

Now we’ve seen what happens when we’re not cheating by using a portion of the dependent variable as an explanatory variable. So let’s start cheating and add in the ENSO data.

chylek emulations GHG AER VOLC SOL ENSOFigure 6. Uses all of the GISS forcings plus the ENSO cold tongue index.

As I said, I couldn’t quite replicate their 0.88 value, but that comes close.

Now, before I go any further, let me point out a shortcoming of all of these emulations in Figs 2 to 6. They do not catch the drop in temperatures around 1910, or the high point around 1940, or the drop from around 1940 to 1970. Even including all of the forcings, and (improperly) giving them different weights, Figure 6 above still shows these problems.

However, all of these global average temperature changes are clearly reflected in the corresponding temperature changes in the North Atlantic ocean … take another look at the bottom right panel of Figure 1. And so of course when they (improperly) include the AMO as an explanatory variable, you get a much better adjusted R^2 … duh. But it means nothing.

chylek emulations GHG AER VOLC SOL ENSO AMOFigure 7. Emulation using all of the variables, including the Atlantic Multidecadal Oscillation (AMO).

PROBLEM THE FIFTH

All of the above is made somewhat moot by a deeper flaw in their analysis. This is the lack of any lagging of the applied forcings. IF you believe in the forcing fairy, then you have to believe in lags. Me, I don’t think that the changes in global average temperature are a linear function of the changes in global average forcing. Instead, I think that there are strong emergent temperature regulating mechanisms acting at time scales of minutes to hours, largely negating both the changes from the forcing and any associated lags. So I’m not much bothered by lags.

But if you think that global average temperature follows forcing, then you need to do a more sophisticated lagged analysis involving at least one time constant.

CONCLUSIONS

• I find the analysis in Chylek 2014 to be totally invalid because they are including parts of the dependent variable (ENSO and AMO) as explanatory variables. Bad scientists, no cookies.

• As is shown by the examples using either GHGs or aerosols plus volcanoes (Figs. 3 & 4), a good fit and an impressive adjusted R^2 mean nothing. We get equally strong and nearly indistinguishable results using either GHGs or aerosols. This is an indication that this is the wrong tool for the job. Heck, even a straight line does a reasonable job, R^2 = 0.81 …

• Giving different weights to different kinds of forcing (e.g. volcanic, solar) is a novel procedure that requires strong physical justification. To the contrary, they have not provided any justification for the procedure.

• As you add or vary the explanatory variables, their parameters change. Again, this is another indication that they are not using the right tool for the job.

• The lack of any consideration of lag in the analysis is in contradiction to their assumption that changes in the global surface temperature are a linear function of changes in global average forcing.

Best to everyone,

w.

De Rigeur: If you disagree with something I or anyone else says, please quote their exact words. That way, we can all be clear on exactly what you are objecting to.

LINKS:

Chylek Paper

Bob Tisdale’s Analysis

Nic Lewis’s analysis

DATA:

GISS Forcing

CTI

GISS LOTI

AMO

Advertisements

  Subscribe  
newest oldest most voted
Notify of
TRM

Welcome back Bob. Nice work.

TRM

Doh. Too early in the morning. I see it is Willis, still nice work 🙂

Accidentally including a dependent variable is a terribly easy mistake to make.
I did this once by accident when calculating FX rates for a merchant bank. FX rates are all highly dependent on each other, they are tied to interest rates in the economies of the various currencies. The issue was I accidentally included part of the result of the previous calculation in the new value, instead of calculating everything cleanly.
The result was chaos – literally. I had constructed a random number generator. But it was very subtle chaos, it looked like everything was working – unless you entered a radically different value for the new FX rate, in which case the kink caused by the previous calculation became painfully obvious.
So like I said – a terribly easy mistake to make. Still, kind of careless for trained scientists to put in a peer reviewed paper. Hopefully there will be a swift retraction.

See - owe to Rich

Willis,
I agree with your Problem the First part one, namely inclusion of the AMO, but I partly disagree with your Problem the First part two, namely the El Nino variable used. My reason is that rather than adding global SST effectively, they are subtracting it (as an anomaly). So I don’t think they can be criticized for the (negative) global SST part. Possibly they can be criticized for using the temperatures in that El Nino rectangle of the Pacific, but it is not clear. It appears that what they are trying to do is cancel out the effect of the SSTs in that rectangle in order to see if they have a knock on effect on land temperatures, which it is generally thought they do.
For this effect-of-El-Nino on land temperatures, what variable would you have preferred? Perhaps just using the SOI index alone, which is based on pressures rather than temperatures?
Rich.

RichardLH

Well done Willis. A nice example of how to get the ‘right’ answer by including the answer itself in the reasoning.
I also think that assigning different weightings to different factors without explaining why the different weightings are needed to how to derive them is very poor science.

Robin Edwards

Thanks for this, Willis. I am intrigued, to say the least. I’ve looked superficially at the data sets you listed, and find, as so often, that the data columns are not headed by a column name. I wonder why this is so, since without this, or a clear and direct route to it, the data are from many aspects useless!
Having written a software package myself, which I sold successfully for quite a number of years and which was based originally on regression analysis, I believe that I have some feel for what has (and has not) been going on in the analyses by Chylek et al. You have clearly pointed out the amateurish or cavalier (or so it seems to me) approach they have to selection of appropriate “independent” variables. It is astonishing! The first question is are they really experts in multiple regression? What software have they used. What regression diagnostics are incorporated in that software? Are they familiar with the concepts of Variance Inflation Factor and the related multiple correlation coefficient for the independent variables? Are they aware of influential points? Are they aware of Cook’s distances? Has a professional statistician whose specialty is multiple regression vetted the paper, either at the writing stage or during “Peer” review? If the peers do not include someone with appropriate background the paper should be viewed with great suspicion.
I would like to try some analyses of the data they actually put into their regression analyses, which may be different from those in the data sets you referenced. How can I be sure what they used? Can you help me with this, please? I would be delighted to hear from you.

daddylonglegs

A big thanks to Willis for his fine job of eviscerating Chylek.
What do his spilled entrails tell us about coming climate? More no doubt than his paper.

Steve Keohane

Thanks Willis, looks good.

daddylonglegs

As you add or vary the explanatory variables, their parameters change. Again, this is another indication that they are not using the right tool for the job.
This important observation points to the fundamental nonlinearity of the climate system being studied, as Willis alludes in the closing sentences of this fine analysis.
As was nicely articulated by James Gleik in his book “Chaos”:
Nonlinearity means the act of playing the game has a way of changing the rules.
Nonlinearity is why this entire approach of simple arithmetic wirh “forcings” is bogus. It ignores the biggest forcing of them all – the system itself.

Paul Vaughan

“Heck, even a straight line does a reasonable job, R^2 = 0.81”
A straight line fails diagnostics. (Does anyone bother with diagnostics??)

Global cooling

What you say about variables is OK, but there is some truth behind. Temperatures of the atmosphere follow the temperatures of the ocean waters. Mixing these waters result in the weather and also the climate that we are talking about here. Showing that IPCC:s forcings does not play a big role is also a result that I like to have on (a peer reviewed) paper.
As the others already said, linear correlations are not enough to explain non-linear systems with feedbacks and lack times.

Crispin in Waterloo

Willis, I side with Rich on the explanation of what they were trying to do in problem one, part two, but with you on the issue of validity.
I have been facing for some time a problem with trying to deal with the outputs from a method of analysis popular in the USA that has metrics (used to express product performance) that are not grounded in physical relationships. The paper seems to suffer from very similar things though the ways it is discussed are different.
It is laborious but if you were to write down the whole equation for the output it is then possible to point to elements of it which are incorrectly repeated. The problem I face is they divide by an independent variable to produce a (claimed) “specific” value determined from a whole value. You are showing that they are using a dependent variable to produce the whole value. Both processes give a result that is ‘a number’ but not one with any valid meaning. I commend you for highlighting the error.
There are several levels of peer review which might be applied to the paper. The first is a review to see that the units are correct and consistent. The paper fails on that score for usings different weightings without explanation.
The next is a logical analysis which is what you have applied to identify problems that can slip through a check of the units. The calculation chain contains illogical steps.
Robin Edwards provides a third level of review which is to look at whether the correct ‘standard’ analytical tool has been selected for suitability and then correctly applied. Excel famously does some stats calculations incorrectly so while everything else might be right, the tool might be too blunt to work the material.
Here’s an analogy incorporating your observations (Chylek’s errors): My ability to predict the winning 7 numbers in a lottery is greatly enhanced if I have access to the winning numbers and can incorporate them in some statistically weighted manner into my number selection algorithm. I then claim my pick-7 method is better than a random guess.
Ya think?!?

ferdberple

so forcing + temperature does a better job of explaining temperature, than forcings alone. what will climate science discover next? that water is wet?

The first is a review to see that the units are correct and consistent.
================
In physics that is a vital step. making sure you add units to every term, to check that the units of the result are in fact what you are calculating. otherwise you can end up with apples + oranges = grapes.
yet in this paper the authors appear to have combined terms with all sort of different units, without regard for the fact that their is no mathematical significance to their combination. you need a conversion table to combine different units, which they have “invented” by using parameters to combine the terms. however, this conversion table changes for calculation to calculation. it is a physical nonsense. the correlation is a spurious artifact of the method.

Kristian

Willis, you write on your first problem: “Then there are pressure-based indices like ENSO, based on the difference in pressure between Tahiti and Darwin, Australia.” This should be SO (Southern Oscillation).

Kristian

The main problem with ANY so-called ENSO index is that it never incorporates the entire actual physical process that is ENSO. It concentrates on one part only (mostly the eastern one). Doing regression analysis on such an index and then conclude something about the global influence of the ENSO process shows a lack of understanding of how the Earth system operates. Yet EVERYONE does it. And that’s why no one still doesn’t seem to get how ENSO affects Earth’s climate.

Richard M

What if the AMO (or pick any ocean cycle) really does drive the global temperature? How do you decide that is the case? I don’t know that including it is 100% wrong even though it is dependent. It’s kind of a catch 22 situation.
In addition, when you detrend the ocean data you require another forcing like GHGs. What if the forcing really is the ocean cycle itself? This analysis just arbitrarily eliminates it. Is there a way to know the right way to approach the problem? Probably not with the limited data we have today.
BTW, I suspect you could do an PDO + Solar + Volcano evaluation and get something with just as good r^2.

Max

“ABSTRACT: A multiple linear …”
They lost me right there, within 3 words.

philincalifornia

“…… using the known radiative forcing …..”
Do they supply a number for the known radiative forcing of ~400ppm CO2 on a background of 30 – 40,000 ppm of water vapor in 2012 minus the known radiative forcing of ~280 ppm CO2 on a background of 30 – 40,000 ppm of water vapor in 1900. Further, how does the difference heat the water ?

Ralph Kramdon

We don’t need a correlation to predict past temperatures, we know what the temperatures were. The only purpose a correlation serves is to predict future temperatures and you can’t do that if you use a dependent variable as an input. I don’t know what the authors were trying to do.

DocMartyn

“I find the analysis in Chylek 2014 to be totally invalid because they are including parts of the dependent variable (ENSO and AMO) as explanatory variables. Bad scientists, no cookies”
I agree, BUT, it is quite clear that the AMO and ENSO exist. Could you live with a ‘graphology’ based sine wave that simply states we have heat sloshing around the globe with a periodicity of about 6-years that raised the Northern Hemisphere by +/- 0.2 degrees during the course of its cycle?

Bill Illis

The Raw Undetrended AMO timeseries has a very interesting relationship to Hadcrut4 temperatures back to 1856.
http://s24.postimg.org/thoyqbbnp/Hadcrut4_vs_Raw_AMO_Jan14.png
Something is causing these up and down swings. True, the AMO region is about 8% of the planet’s surface, but why is it so representative of the planet’s overall up and down swings? 8% –> 100%.
The Nino 3.4 Index region is only 1.2% of the planet’s surface and global temperatures seem to lag behind it by 3 months to the tune of +/- 0.25C. 1.2% –> +/- 0.25C which makes it a big driver of natural climate change.
The other issue is, the North Atlantic does have swings in its underlying Ocean Heat Content. One would not expect the deep ocean down to 400, 700 and 2000 metres to respond so quickly to surface temperature variations.
There is an underlying mechanism here. The Raw AMO versus Ocean Heat Content back to 1955. (100, 700, 2000 metres)
http://s28.postimg.org/p67iurpx9/Nor_Atl_OHC_vs_Raw_AMO_2013.png
And then the North Atlantic temperatures down to 400 metres going back to 1900 (with some years missing). The Blue and Red lines here which cover most of the North Atlantic. [Note, this is probably the only place where we have deep ocean temperatures going back this far].
http://s14.postimg.org/dq96zg4b5/North_Atlantic_Temp_400m_01.jpg
I’m okay with the AMO being a real natural cycle that has an underlying ocean circulation system oscillations which impact the global climate by as much +/- 0.30C (more than the ENSO in fact). It should not be ignored.

Steve from Rockwood

Nice analysis Willis. I continue to wonder why climate scientists use linear regression to explain temperature changes. I would prefer they express the forcings mathematically and subtract them one by one from a non detrended temperature data set that has not been tampered with. What is left over is what they don’t know. And I suspect it would be mostly leftovers.

dudleyhorscroft

Reckon you could get a pretty good R^2 value from two variables — a linear increase of 0.8 K per century (commencing at -0.3 in 1910), plus a saw tooth profile with period 60 years, comprising 40 years from -0.1 to +0.1 K then 20 years of +0.1 K down to -0.1 K, with the first minimum at 1910. Look at your Fig 2, and see what happens.
Better R^2 than the straight line you have shown? But what the independent variables are, I have no idea, and see no reason for them, except – “We are getting into a warmer bit of the Galaxy, plus the Sun is a very long term Cepheid Variable.”

Kristian

Willis, you say: “I find the analysis in Chylek 2014 to be totally invalid because they are including parts of the dependent variable (ENSO and AMO) as explanatory variables. Bad scientists, no cookies.”
I agree with you on the AMO part. I do not agree with you on the ENSO part. The ENSO process is what drives the global warming (through among other things the AMO) over multiple decades like 1976-2001. There is no ‘external’ warming driving ENSO. ENSO is what let’s the heat in the first place.
There is however ‘something’ modulating the effect the ENSO process has on global climate, switching between a warming and a cooling mode on multidecadal time scales. This ‘something’ seems to involve the tightly coupled ocean/atmosphere connection in the Pacific, like when the SOI stepped down over night in 1976/77 and stayed there for the next 30 years.
You might very well be correct in your hypothesis that the switch occurs whenever the temperature gradient between the Equator and the Poles is getting either too steep or too gentle. I don’t know. But ENSO is not ‘dependent’ on some ‘background’ warming to work. The ENSO process is what ‘creates’ the warming, building OHC, SST and tropospheric temps globally.
The energy comes from the Sun. ENSO, however, is the grand distributor/administrator of this energy.

Kristian

“ENSO is what [lets] the heat in the first place.”
And by that I meant letting the ‘the extra heat’ in creating general warming. ENSO also determines how much of the absorbed heat it is willing to let out again.

Theo Goodwin

Your “nuts and bolts” analysis is very useful and very important. You have highlighted what I take to be the main issues in a way that is reasonably clear to anyone who cares to pay attention. Thanks much.

Willis writes: “Note that the R^2 of a straight line is quite high, 0.81. So their correlation of 0.88 … well, not all that impressive.”
Thanks.

Dudley Horscroft says:
March 16, 2014 at 7:59 am
Reckon you could get a pretty good R^2 value from two variables — a linear increase of 0.8 K per century (commencing at -0.3 in 1910), plus a saw tooth profile with period 60 years, comprising 40 years from -0.1 to +0.1 K then 20 years of +0.1 K down to -0.1 K, with the first minimum at 1910. Look at your Fig 2, and see what happens.
Better R^2 than the straight line you have shown? But what the independent variables are, I have no idea, and see no reason for them, except – “We are getting into a warmer bit of the Galaxy, plus the Sun is a very long term Cepheid Variable.”

Your wisdom and humility by acknowledging “We don’t know the specific cause, duration or number of simultaneous climate cycles going on at the same time” is reasonable. The only real answer actually.
However! And you knew there was a “but” coming, didn’t you? The long range climate is most definitely NOT linear, so NO LINE of ANY SLOPE will ever be correct for longer than 25 years!
But, let’s assume we see a 1000 year climate cycle, a 200-210 year climate cycle, and a 60-63 AND a separate 68-69 year ENSO cycle. (The 68 year cycle tracks very, very well based on fisheries proxies in the Pacific going back to the mid 1600’s.) Now, look again:
Is there a “separate” 1000 year cycle (of perhaps 0.1 degree amplitude) PLUS 63 year cycle PLUS a 68 year cycle PLUS a 200 year cycle? Or do we now see a combined peak of 3×68 cycles (204) coming to a peak with 3×60 year cycles ? Does the 1000 year cycle happen because of resonance between 5x 200 year cycles and a 63 and 68 year cycle?
Just because we are measuring a slow rise of temperatures superimposed with a 60-some-odd year short term oscillation since 1650 to 2000, and a pause from 1997 through 2014, doesn’t mean temperature won’t drop again slightly, then peak around 2060 … Then slide down towards the next “Modern Ice Age about year 2350. Or maybe 2410.
The next Nobel Prize needs to go not to the politicians killing people with their propaganda about climate cycles, but to the scientists who DO figure out WHAT those cycles are, and WHY those cycles repeat themselves through history.

The solar forcing used by Chylek is very likely not correct. There is no evidence for the sharp increase in solar activity since 1900, followed by near constancy [apart from the cycles] after ~1950.

George Turner

Assigning different weights to the W/m^2 forcings is really no different than expressing the forcings in different units, with some in degrees F, some in calories, and some in BTUs/hr, kJ/kg, etc, and then proceeding to plug the values straight into some equation.

Martin Lewitt

Assigning different weights to different W/m^2 forcings is actually a refreshing admission of complexity and nonlinearity. Each of the forcings coupled to the climate differently in vertical and geographical distribution and in some cases chemically (as in Solar generating ozone), etc. In a nonlinear dynamic system it is the assumption that they were all equivalent that would have to be justified. Representing each forcing by the variation in a globally and annually averaged W/m^2 figure is a poor proxy for these coupling differences, but then a grid from the ocean mixing layer to the stratosphere is the reason we have AOGCMs. The reason the author didn’t explain the allowing of different weights is that it is common knowledge, e.g., here are other refreshing acknowledgements of the implications of non-linear dynamics. Knutti and Hegerl state in their 2008 review article in Nature Geoscience:
“The concept of radiative forcing is of rather limited use for forcings with strongly varying vertical or spatial distributions.”
and this:
“There is a difference in the sensitivity to radiative forcing for different forcing mechanisms, which has been phrased as their ‘efficacy’”

taxed

As l have hinted at in other posts, what looks to me to be a key factor in major climate change in the NH is the Azores high pressure pattern. Its the fact that its a fairly stable pressure pattern is what is important. Because with the clear sunny skies under it means there is ready supply of warm water there to be taken up by the wind driven Gulf Stream towards europe. But what’s more important is the Azores high often ridges up towards europe and draws warm air up into europe, so giving europe a much milder winter climate then would other wise be the case.
And its this block of warmth over the northern Atlantic and europe the Azores high causes is what is so important. Because it stops the bitter Arctic winter weather that forms in northern America and northern Asia from meeting up to become one huge block of cold weather spreading over the NH. Just like it did during the last ice age.

Gary Pearse

My problems start with the GISS forcing diagrams as you have pointed out with Ozone. Man there was a lot of black carbon when I was a kid in the 40s and 50s when I had to shovel coal into a leviathan hot-air octopus down our basement in winter time in Winnipeg. My father was also a steam locomotive engineer who put who burnt 16-18 tons of coal each way per division of 130 miles. Pretty much all homes were heated with coal or wood in those days.
And are they admitting to a rising hockey stick in snow albedo? Shouldn’t they have this basckwards?

Willis Eschenbach

Paul Vaughan says:
March 16, 2014 at 5:28 am

“Heck, even a straight line does a reasonable job, R^2 = 0.81″

A straight line fails diagnostics. (Does anyone bother with diagnostics??)

Thanks, Paul. Of which diagnostics are you speaking?
w.

Keith

RACookPE1978 says:
March 16, 2014 at 8:29 am
The next Nobel Prize needs to go not to the politicians killing people with their propaganda about climate cycles, but to the scientists who DO figure out WHAT those cycles are, and WHY those cycles repeat themselves through history.

I’d hope that those scientists would pick up Nobel prizes for Physics, Chemistry and possibly Biology, which may play some part along the line. In other words, the ones worth winning that are presented in Sweden. No doubt Albert Gore IV would still be picking up the Nobel Peace Prize in Oslo for his work in shopping his heretic school science teacher to the CIA (Climate Inquisition Authority).

Willis Eschenbach

lsvalgaard says:
March 16, 2014 at 8:42 am

The solar forcing used by Chylek is very likely not correct. There is no evidence for the sharp increase in solar activity since 1900, followed by near constancy [apart from the cycles] after ~1950.

Thanks, Leif. In addition to the issue you correctly raise, there are a number of other problems with their underlying forcing datasets … as I said, does anyone believe that the ozone went flat in about 1970? I didn’t want to touch any of those issues, as I already had five problems under discussion. However, they are serious issues and worthy of discussion..
w.

Willis Eschenbach

Martin Lewitt says:
March 16, 2014 at 9:33 am

Assigning different weights to different W/m^2 forcings is actually a refreshing admission of complexity and nonlinearity. Each of the forcings coupled to the climate differently in vertical and geographical distribution and in some cases chemically (as in Solar generating ozone), etc. In a nonlinear dynamic system it is the assumption that they were all equivalent that would have to be justified.

My point is that the various values are all in watts/m2, and they are all added together by the GISS and IPCC folks to give us a “total forcing”. As a result, the procedure that Chylek et al. use is not valid GIVEN THE ASSUMPTIONS OF THE IPCC.
As to what’s happening in the real world? Mainstream climate science doesn’t deal with that …
w.

ren

The surface temperature of the oceans in winter is strongly dependent on wind direction (polar vortex).
http://weather.unisys.com/surface/sst_anom.gif

ferd:
“yet in this paper the authors appear to have combined terms with all sort of different units, without regard for the fact that their is no mathematical significance to their combination”
remember that when you ….
A) look at cycles papers from sun nuts
B) read mckittrick and micheals (2007 and 2010) on UHI.

“I’m okay with the AMO being a real natural cycle that has an underlying ocean circulation system oscillations which impact the global climate by as much +/- 0.30C (more than the ENSO in fact). It should not be ignored.”
it shows up in the residuals which is probably the right of doing the analysis

“Assigning different weights to different W/m^2 forcings is actually a refreshing admission of complexity and nonlinearity.”
you should also note that when people combine them into one lump skeptics howl.
They howl if you combine them. they howl if you separate them.
must be the moon

DR

I think it is synchronized chaos, and nobody alive today understands even a small portion of how the climate system works.

Steven Mosher says:
“…when people combine them into one lump skeptics howl. They howl if you combine them. they howl if you separate them.”
I notice you are lumping all skeptics together.
Must be the moon.

Crispin in Waterloo

When skeptics tear a dreadful paper apart for theoretical and analytical reasons, people post vacuous half-breaths of diversion. When skeptics tear a dreadful paper apart for practical and evidentiary reasons, people post vacuous half-breaths of diversion.
Must be the M☺sh.

Crispin in Waterloo

@ferdberple
>>The first is a review to see that the units are correct and consistent.
================
>In physics that is a vital step. making sure you add units to every term, to check that the units of the result are in fact what you are calculating. otherwise you can end up with apples + oranges = grapes.
It happens that I was taught that by a physicist. A proper one. 🙂
>yet in this paper the authors appear to have combined terms with all sort of different units, without regard for the fact that their is no mathematical significance to their combination.
The reduction of forcings to Watts/m^2 is sort of an attempt at universalizing disparate factors, but forcing is not that simple. There is a time element and a cumulative effect aspect and where was that factored in? Willis correctly points out that the IPCC has a different approach.
>you need a conversion table to combine different units, which they have “invented” by using parameters to combine the terms. however, this conversion table changes for calculation to calculation. it is a physical nonsense. the correlation is a spurious artifact of the method.
There is a real chance that it is spurious, some portion of the proof being that several other methods of processing parts of the same data provide nearly the same result. But…as soon as the whole process fails for one logical reason – using dependent variables to calculate the (dependent) output – there is little need to keep re-analyzing it.

TheLastDemocrat

Wikipedia has a decent entry on the variance inflation factor – a well-recognized measure of the degree that collinearity is affecting a multiple regression model.
Eric Worral’s note above is a great real-world example of including redundancies in a regression; the regression weights assigned by the program will be “chaotic,” since almost any estimate will serve for at least one of the predictors in a regression with too much collinearity.

ren

It should be obvious that the regions with the greatest variability in temperature will have the greatist effect on the variability of the calculated global average. They would have done better by testing their supposed forcings on these highly variable areas and not including supposed sensitivities. I can get better R squares with fewer degrees of freedom by curve fitting.

Paul Linsay

The correct spelling is Ouroboros, as in the self eating snake. http://en.wikipedia.org/wiki/Ouroboros
[Fixed. Thanks, -w.]