Well done Martin A you have handled all the spin fro EM with consulate ease.

Radical Rodent
Despite my continued efforts two of my three sons write "would of" instead of "Would have" or even "would've". Mrs S reads all texts and WhatApps as if the were written without typos, bad spelling, bad or non-existent punctuation. My advice to you is the same as I give her, don't worry about it and assume that the writer isn't trying to wind you up.

Martin A
Agree about Entropic being generally polite, one can let a bit of frustration go by. When it is one against several I'm never sure if difficult questions are ignored or left behind as the conversation moves on.

"I find it difficult to reconcile my own experience with your description of the collected climate data as non-gaussian"

Well strictly, I didn't actually say it was non-Gaussian - I said it wasn't independently distributed, zero-mean, constant-variance Gaussian. Failure of *any* of the conditions invalidates the method. Although technically the data *isn't* Gaussian, over short periods it looks close enough not to be an issue, the main issue is actually the independence assumption.

However, it's a more complicated question than it seems, because of the issue of approximations.

But on the question of whether the data itself is trend+Gaussian - consider that if it was, then extending the trend into the past or future would continue to see 99% of the values within 3 standard deviations of the line indefinitely. We can say immediately that it's a physical impossibility since we'd eventually get temperatures below absolute zero, but we don't actually have to go back anywhere near that far before the data diverges.

The answer to that, of course, is that nobody seriously believes that the underlying physics actually predicts "linear trend + Gaussian" behaviour. The conceptual model behind it is that it is actually "deterministic curve + Gaussian" and the linear trend is just the first-order approximation to the curve. What we're saying is that the time period is short enough for the short segment of the curve we can see to look straight. However, it's only an approximation. Where the curve deviates from straightness, the distribution doesn't fit.

We can solve that problem by using a higher-order approximation and fitting to a polynomial. A constant is a degree 0 polynomial, a linear trend is a degree 1 polynomial, we're just extending the same method to take more terms into account.

But how many should we take? If we take as many terms as we have data points, we can get the line to go through all of them exactly, and the apparent error is zero! The problem is that we cannot, by any purely statistical means, distinguish noise from curve. Any wiggle you see might be a kink in the underlying curve, or it might be just noise.

So behind the choice of a linear trend, rather than any higher-order curve, is *another* implicit assumption: that the curve is sufficiently 'smooth' for this to be a good approximation over the selected time interval. More technically, we look at the frequency spectrum of the climate 'signal' and the weather 'noise' and presume that the climate has no power at frequencies higher than a certain level, and the weather has no power at frequencies lower than that level. Smooth the data with a threshold at that cut-off frequency, and you supposedly separate signal from noise. The use of a linear trend implies a cut-off frequency corresponding to about half the interval covered by your data.

However, there appears to be no objective criteria on which such a threshold can be set. The frequency spectrum of the 'weather' appears to merge smoothly with the 'climate'. This is reflected in the correlation properties of the 'noise'. The residuals are not statistically independent, month to month, implying that there is more power at the lower frequency end of the spectrum. Moreover, it keeps rising all the way to the cut-off.

So what does the very low frequency part of the 'weather' spectrum look like; the part of the spectrum where we expect to find the climate change signal? Nobody knows for sure. If it's very strong, then all the rises and falls we're seeing might be just natural background variation. If it decays away beyond a certain point and is weak, then on timescales longer than that level we would be able to detect climate change. What you see in the data depends on the underlying assumptions of your analysis method. If you assume a trend, you see a trend. If you assume no trend, you see no trend. The analysis is simply reproducing your starting assumptions - the conclusions don't arise from the data. It just means we don't have enough data to tell, yet. But if you're *unaware* of the implicit assumptions underlying your analysis method, you could easily get a situation where different people are analysing the same data by different methods and coming to diametrically opposed conclusions, both sides utterly convinced they're right!

"You have that. CMIP5 is a physical model used to project future temperatures."

Quite right. The CMIP5 models can indded be considered to constitute a statistical model of the background. The problem comes when you try to validate it.

Consider the SI from the paper "Quantifying the likelihood of a continued hiatus in global warming" Roberts et al 2015.
http://www.nature.com/nclimate/journal/v5/n4/extref/nclimate2531-s1.pdf

This shows a figure S1 in which the probability of a pause of various lengths and strengths occuring in each of the CMIP5 models is shown (those models not invalidated by their failure to simulate El Nino, that is). If you plot the point corresponding to the current pause (or for that matter, the previous one from 1950-1975), you find almost all of the individual models are rejected at the 95% level by a 15 year pause; indeed many are rejected at the 99% level.

It doesn't make sense to use the full ensemble of CMIP5 simulations for our distribution since many of these are from models we know to be incorrect. It might make sense to use the ensembles from that subset that hasn't been rejected by its conflict with observation, though.

The paper tries to look on the bright side, and suggests that the probability of the pause continuing for another 5 years given that it has already lasted 10 is about 25%. That is to say, even the models think a 5 year continuation is quite likely - presumably the unspoken implication they hope to draw being that if the pause *does* continue we shouldn't therefore reject the models.

This isn't actually at all surprising. If the probability of a pause in each 5 year period was independent (I'm sure it isn't, but let's keep it simple) and about 25%, then the probability of a 10 year pause would be 6.25% and that of a 15 year pause would be about 1.6%, which is pretty much what the models give. So this doesn't give us any reason to think the evidence of another 5 years of pause is any less just because it comes at the end of 15 year pause.

"I note that after meeting Doug Keenan the Met Office have decided to go with their own advisers."

Actually, the Met Office (eventually) conceded that Doug was absolutely correct about his statistical significance calculations, but changed their story to say they had based their significance conclusion on other arguments - based on physics rather than statistics.

The Met Office said:
"Work undertaken at the Met Office on the detection of climate change from temperature observations is based on formal detection and attribution methods, using physical climate models and not purely statistical models, as discussed in Chapter 9 of the Contribution of Working Group I to the IPCC’s Fourth Assessment Report, 2007."

The IPCC of course say in chapter 9:
"The approaches used in detection and attribution research described above cannot fully account for all uncertainties, and thus ultimately expert judgement is required to give a calibrated assessment of whether a specific cause is responsible for a given climate change."

So it all boils down to "Because we say so."
:-)
--

Incidentally, they do give quite a nice summary of the need for a statistical model in the IPCC chapter 9 a few paragraphs earlier.

"Both detection and attribution require knowledge of the internal climate variability on the time scales considered, usually decades or longer. The residual variability that remains in instrumental observations after the estimated effects of external forcing have been removed is sometimes used to estimate internal variability. However, these estimates are uncertain because the instrumental record is too short to give a well-constrained estimate of internal variability, and because of uncertainties in the forcings and the estimated responses."

Quite so.

-----

"Add to that the latest link from NCC 1701E, which shows that “greenhouse effect” does not exist, and you can chuck all of the fancy formulae being generated, and ideas to “mitigate” (usually at horrendously extreme cost), into the bin."

The greenhouse effect *does* exist. But you can't argue with someone who uses a different version of physics to the rest of science, so we all got bored and gave up.

" The point I was trying to make was that the effect of “greenhouse gasses” does not really exist"

Actually, that the problematical bit.

The physics I'm talking about is the bit that says an object in the vacuum of space adjusts its temperature until the IR-visible surface radiating to outer space radiates away exactly as much energy as the object absorbs. If that surface is the solid ground, then the solid ground approaches that temperature irrespective of how much atmosphere is above it.

And if the solid ground is warmer than that, it can only be because the radiating surface is elsewhere - i.e higher up in the atmosphere. And that can only happen if the atmosphere is at least partially IR-opaque.

But we've said all this dozens of times before. There's not much point in going over it again.

Martin A

Despite geronimo's foolish congratulations, you admitted defeat when you stopped arguing the topic and used an as hominem attack instead.

Nullis in Verba

Nice to see someone moderating one of the pseudoscientists.

Which is the tail and which is the dog?

Formal logic is a tool used by philosophers to analyse the structure of arguments. It shares the inflexibility of other logical systems.
The problem with logical systems is that depending on ones axioms one can "prove" anything. Read Kaballah as an example.

Statistics are analytical tools used to assess data. The statistics does not supersede the data, especially not in the sense you are presenting them.

In both cases they do not reflect that science as it is practiced uses a much fuzzier logic.

RR you got me there'd I spelt consummate with one "m" and the spell checker took over. Still everyone makes mistakes, that's why they put rubbers on the end of pencils😀

"In modern English, the plural of gas is usually gases, and gasses is the simple-present verb. For instance, we might say that the more Bill gasses up his car, the more greenhouse gases his car emits."

EM. Why don't t you put your fuzzy logical abilities into what blunder Dana, and his imitators made in the 1910 - 1940 temperature calculations? Although to be honest regular logic, or just plain common sense would do the trick.

"So many of his witnesses observed the utter freedom of his flights of thought, yet when Feynman talked about his own methods not freedom but constraint ... For Feynman the essence of scientific imagination was a powerful and almost painful rule. What scientists create must match reality. It must match what is already known. Scientific creativity, he said, is imagination in a straitjacket."

"...imagination in a strait jacket." Doesn't like a proponent of fuzzy logic in science to me. "