“The IPCC is all for testing the models”

Then why haven’t they done so publicly? All we got was chapter 8 with no validation what so ever, no statistical measures, not even residuals plotted. What they have shown us is NOT “testing models” if by that you mean verification and validation.

And what about other ways to test models? E.g. Douglass et al IJC 2007? ( http://www.nsstc.uah.edu/atmos.....s_etal.pdf ). Or the complete lack of periods with decreasing heat content in models (except for major volcanoes) when we can clearly see such a period happening right now?

I am not suggesting that ENSO and the solar cycle are larger than previously thought. The solar cycle references I’ve found indicate roughly 0.1C peak-to-trough.

The modellers insist that the solar effect is entirely TSI and we know that TSI cannot be used to explain a 0.1 degC amplitude cycle. This means you are assuming that some non-TSI effect is at work. Also analyzing the data to determine the magnitude of the solar effect and the ENSO effect is an attempt to apply the same correction twice because it is not possible to seperate solar from ENSO in the raw data.

Lubos’s example is relevant because it illustrates how small changes in amplitude between cycles can produce a trend. We have no reason to beleive that either the solar cycle or ENSO are perfectly symmetric phenomena that would cancel out over time. Therefore, any analysis that is presumes that they cancel out over time is flawed.

Lastly, have you ever used a hot plate? They work by adjusting the duty cycle of an electrical pulse. Increase the duty cycle and the plate heats – decrease and the plate cools. This is a pretty simple mechanism that could explain a trend induced by solar cycles of different length. Schwartz’s paper an Lucia’s subsequent analysis demonstrate that the time constant of the earth system has the right order of magnitude to make such a mechanism plausible.

I am interested in the truth as well – whatever it may be. However, I don’t believe that we know enough about what is going on to justify the a priori assumption that nothing other than CO2 can be the source of a long term trend.

I am not suggesting that ENSO and the solar cycle are larger than previously thought. The solar cycle references I’ve found indicate roughly 0.1C peak-to-trough. Lucia has done her own ENSO analysis. We are not attempting to emprically determine the weather noise. These are all attempts at figuring out the truth. Is that your goal as well? Jump in and help out.

Lubos’ analogy of adding and removing 1% per month does not make much sense. The bi-monthly removals are larger than the additions because the removals include the previous additions. Try adding and removing $1 every month instead of 1%.

Your comparison to vibration of mechanical systems is not valid for temperature. Resonance does not exist for heat.

Given the trend structure identified by John V, 0.1 must be a lower bound

John V’s appeals to ENSO and the solar cycle are two edged swords because they must have an effect that is much larger than previously claimed in order to explain the variance. This larger effect would have to be subtracted from warming that was previously assumed to be due to CO2.

More importantly, these cyclic effects are not symmetric and poorly understood which means no one can reasonably claim that they cancel out over the long term. Lubos gave a good analogy to illustrate the point: how much would you have in a bank account after 10 years if the bank added 1% one month and then deducted 1% the next? (answer: a net decrease of 1% over the ten years).

Another analog to consider: a system driven with a periodic signal will accumulate energy if the frequency of that signal is faster than the time constant for the system. IOW a few short solar cycles could cause a upward trend if the solar period is shorter than the time constant of the climate system (which it appears to be the case). This accumulation would be insignicant if the solar effect is small (which is the original position taken by the modellers) – it becomes significant if the solar effect is actually large enough to explain the variance from the IPCC trend.

In short, I don’t think we can make any assumptions about the upper or lower bound on the real CO2 induced trend. However, the data to date suggests that is mostly likely lower than higher.

The full set of global near surface temperature anomalies is now available for April 2008 with the release of HadCRUT3 saying +0.25°C (decline from +0.43°C), have you updated your infamous chart?

Is your spreadsheet for doing this analysis online?

I think John V and I are at least looking at the same hypothesis, and debating the uncertainty intervals. He’s going all Bayesian on me (which is fine). He’s suggested a way to do something. I agree with the concept, but I”m trying to figure out if I agree with the details. I said I’d take a look at it and figure out what I get.

In this respect, John V. seems to believe that his T = A + E + S + O + W model is accurate enough. Perhaps it is. I doubt it. It does look like a good start to tackle the “weather noise” problem, at any rate. But that’s all. And therein lies the problem: overselling of the skill of their models by some climate scientists and the IPCC.

“Too large to say anything useful” seems to me to be the proper conclusion for any short interval, given the obvious variability of trends demonstrated in 2831.

Let me try to summarize the argument, to see if I understand.

1. The SPM says:
Model experiments show that even if all radiative
forcing agents were held constant at year 2000 levels,
a further warming trend would occur in the next two
decades at a rate of about 0.1°C per decade, due mainly
to the slow response of the oceans. About twice as
much warming (0.2°C per decade) would be expected
if emissions are within the range of the SRES scenarios.
The problem is, no confidence bounds are specified.

2. Lucia observes that the recent trend is -0.07+/-0.2 (translating to C/decade).

3. RC elaborates on the SPM with a histogram (reproduced above) that says the 7 year trend is 0.19+/-0.42 (translating to C/decade and using 2x the fitted normal sigma) and the 20 year trend is 0.21+/-0.18 (over a different interval).

4. Lucia argues that RC’s distribution confounds weather and model/parameter variation, and that the real sigma should be smaller.

5. John V shows in 2831 that 7 year trends have large deviations from underlying longer term trends, with an 80% range of -0.18 to +0.3

From 2, Lucia concludes that the central IPCC projection is falsified, because -0.07+0.2=0.13 < 0.2. Three counterarguments emerge in comments:
a. From 3 and 5, the error bars on a short term trends understate true variability. If the error in 2 is a third larger (0.27 instead of 0.2), the IPCC central tendency is not rejected.
b. The 7-year trend and the 20-year IPCC projection are not directly comparable (note the sigmas of .21 vs. .09 in 3).
c. The fact that the central value of a projection without specified confidence bounds lies outside an interval conveys information about that particular value, but not the projection or models themselves.

There are of course counterarguments to the counterarguments, but it seems to me that this all boils down to a question of the proper origins of error bounds used to assess falsification.

Consider a thought experiment, reframing the problem as a Bayesian update. The prior is the IPCC projection, let’s say it’s distributed N(trendP,sigmaP). The observation is N(trendW,sigmaW). The posterior is Normal,
mean = (trendW/sigmaW^2+trendP/sigmaP^2)/(1/sigmaW^2+1/sigmaP^2)
sd = 1/sqrt(1/sigmaW^2+1/sigmaP^2)
trendP = 0.2 (IPCC); trendW = 0.07 (Lucia).

From a modeler’s perspective, it seems obvious that sigmaP should be based on AR4; i.e. you don’t arbitrarily assert that a forecaster’s confidence bounds are narrower than stated (though you might say that they are uselessly wide, which is a different question). Thus one could choose 0.09 for sigmaP (from the RC evaluation of the AR4 ensemble 20-yr prediction). The remaining question is what to choose for sigmaW – 0.1 (Lucia’s estimate) or 0.2 (RC’s AR4 7-yr ensemble). If you pick 0.2, the modeler’s posterior is 0.15 deg/decade – not a big change, consistent with the low information content of the 7-yr data. If you pick 0.1, it’s .08 – a big change, but viable only if you believe that the true temperature trend is stationary.

Personally I’m in the 0.2 camp. Given the trend structure identified by John V, 0.1 must be a lower bound. 0.2, on the other hand, is not necessarily an upper bound; models might understate true variability as well as overstate it (e.g., due to missing dynamics, as with low-order models).

Two other observations: 1. All of this is confounded by measurement error in temperatures. 2. Talking about a summary statistic (trends) is less direct than talking about absolute temperatures or temperature anomalies; a change of variables might clarify some points of the conversation.

Models have been around for no more than 20 years…

You’re off by about 20 years.

Sellers, W.D., 1969: A Global Climatic Model Based on the Energy Balance of the Earth-Atmosphere System. J. Appl. Meteor., 8, 392–400.

You believe that there “would be a rush to explain why the models ‘under-predicated’ the warming”. That might be true. I know with 100% certainty that there has been a rush to show that the models over-predict the warming. Your conjecture is interesting but is a clear example of a strawman argument.

I’m all for testing the models. The IPCC is all for testing the models. As knowledge improves, the IPCC has revised the predicted trend with every new report. I am definitely not saying the models are infallible. All I’m saying is that there is not enough information in the 7-year trend to say much about the models. I am trying to figure out appropriate ways to make the uncertainty interval smaller so that a test can be performed on a 7-year trend. Maybe you’d like to help?

I’m with Boris on this one. If you see any examples of the type of activity you predict might happen, let me know and I’ll argue against them.

Why are you limiting yourself to seven years? Test the models over a longer period–wait, it’s been done and the answer is not what you are looking for.

Hindcasts don’t count. The latest batch of models starts in 2001 – there is no longer period to use. Models have been around for no more than 20 years yet the 1990 vintage models have also overestimated the effect of CO2.

Do you honestly believe that the IPCC and various warmers would have applied that standard if the real data happened to be warming faster than predicted over 7 years?

Maybe some would have. That doesn’t mean it’s right. Call me if you see someone doing this in the future, and I’ll be happy to correct them.

Why are you limiting yourself to seven years? Test the models over a longer period–wait, it’s been done and the answer is not what you are looking for. Now I see why seven years is so important!

Of course I realize such arguments mean nothing to people who think climate models are complete junk.

To you it may mean 7 years. Does the IPCC say anything about 7 years?

Do you honestly believe that the IPCC and various warmers would have applied that standard if the real data happened to be warming faster than predicted over 7 years? If 7 years of data indicated rapid warming there would be a rush to explain why the models ‘under-predicated’ the warming and demands for even more aggressive policy responses. The -its only 7 years- argument is only used when the data does not support the policy objectives of the IPCC.

7 years is long enough to justify asking whether there is a bias in the IPCC models. Blind trust in the model predictions is most definitely not justified given the data so far. If the models are right it will become clear by the next solar max. If they are wrong we could potentially save ourselves a lot of grief if we wait and see. Policy should be driven by the data and the data right now says the models could be wrong.

Of course, I realize that such arguments mean nothing to people who have decided that the models cannot possibly be wrong.

The reason the March anomaly was adjusted downward from 0.67 to 0.60 is explained here. The net is that much of the station data are made up of estimates, and those estimates are based on the average temperature of the month being estimated as well as the average temperatures of the other two months in the same season. As more data is added to what is already there, the averages change and hence, earlier estimates can change.