Up to 2010, Foster, Rahmstorf and Cazenave had the liberty of “taking out” “exogenous factors” (e.g. ENSO, a bogus method but we’ll go with it here) and that worked to their advantage for fixing “the true background anthropogenic global warming signal”. That had the effect of pulling down all levels prior to their pivot point at 2010, a manipulation that has also been going on in major observation series i.e. hindcasting that RF&C Fig 1 trajectory leads to very silly levels prior to 1980:-

http://www.skepticalscience.com/pics/FR11_All.gif

But now that RF&C have immutably fixed the “signal” (according to them), they will NOT be able to pull down their 2010 pivot point level as new data comes in because they no longer have that liberty, the “signal” was fixed in their Fig 1.

And for 2013 – 2017, RF&C will also have to “take out” any El Nino that may occur to be consistent in their method. Any continuation of their “signal” will have to be anthro-only – no El Nino’s allowed.

Neither will they be able to “add in” any imaginary warming after their fixed 2010 pivot point to perpetuate the “signal” if observations after “taking out” El Nino’s do not produce higher levels relative to their 2010 Fig 1 pivot.

Over the next 5 years, Foster, Rahmstorf and Cazenave (and SkS) are about to hoist themselves on their own petard unless there’s some radical warming in that time.

Looks like I’m wrong here. The RF&C Fig 1 trajectory does intersect roughly with the 2016 peak in the UKMO 2012 projection.

So the “true background anthropogenic global warming signal” in both SkS/RF&C and HadGEM3 will be tested over the next 5 yrs.

And if elevated temperature levels don’t eventuate to perpetuate the “signal”, AGW is dead.

>’The models might even improve. But will this change the tune that is sung?”

The acid test will be on this decadal projection over the next 5 yrs given that there is improved accounting for natural variability. But although the trajectory is radically modified downwards from 2011 it is still for “record” levels above 2010′s strong El Nino year level and 1998′s superstrong El Nino level. The record levels do not arise primarily from natural factors though from what I can gather but from the assumed “anthropogenic global warming trend” because there’s no natural factor on the horizon short of a superstrong El Nino to boost temperatures to record levels.

2012 was essentially an ENSO neutral year (a little more El Nino than La Nina) so that temperatures oscillate above and below a 2012 neutral level for this current decadal regime. See RSS (Note the different baseline to the UKMO HadCRUT 2012 projection plot – they’re not directly comparable):-

http://junksciencearchive.com/MSU_Temps/RSSglobe.png

If we take the average of 2012 as roughly 0.2 anomaly, then “record” levels in RSS terms are at least 0.4 higher (2010) or 0.6 higher (1998) than 2012 ENSO neutral levels. The lower confidence level if it eventuates would still be 0.25 higher than 2012.

If this projected boost does not occur over the next 5 yrs, there cannot be a “background” anthro signal of any significance. And say La Nina dominates. That would mean oscillation BELOW 2012 levels i.e. cooling. There’s already a La Nina on the way for 2013 possibly and the PDO is in cold mode so La Nina could dominate for the next 20 yrs plus as La Nina did for the last 30 yrs with PDO in warm mode.

Then there’s the solar recession upon us, that can’t boost warming. It’s only lagged ocean heat that will keep temperatures elevated but wont do any boosting. With solar forcing reduced, temperatures must decline at some point in the future – when?

So if over the next 5 yrs temperatures even just maintain 2012 neutral levels on average, the tune that is sung will of necessity have to change.because the assumed “background” anthropogenic signal demands a rise.even at the lower limit by 0.25 C and at mid confidence by 0.4 – 0.6 C by UKMO’s forecast in RSS terms.

The probabilities that could be assigned to the scenarios would however provide a prediction more like this in my view going by natural indicators (note these are probability factors not anomalies):-

0.2 warming (anthropogenic forcing above neutral levels, El Nino prevailing)
0.3 no change (neutral conditions prevailing)
0.5 cooling (La Nina conditions prevailing, solar recession)

The next 5 yrs are crucial for DAGW.now.

http://newzealandclimatechange.wordpress.com/2013/01/19/the-halt-in-warming/#comment-1155

Unless UKMO is the as yet unidentified model that tracks the land-based observations. It could well be I suppose, given their HadGEM3 development. That work may have been transferred to their CMIP5 model in some way:-

“The Met Office has been the first centre to start producing data and to have finished all the experiments using our state-of-the-art “Earth System” model, HadGEM2-ES”

http://www.metoffice.gov.uk/research/news/cmip5

Decadal Forecasting – What is it and what does it tell us?

Decadal forecasts are designed to forecast fluctuations in the climate system over the next few years through knowledge of the current climate state and multi-year variability of the oceans. This item provides some more detail on what they are, and what they can tell us.

>>>>>>>>

http://www.metoffice.gov.uk/research/news/decadal-forecasting

Figure 5 is rubbish because the blue 2012 spaghetti has been placed far too high compared to Figure 1 and other synthesis plots elsewhere.

So if UKMO had to develop HadGEM3 in order to realistically mimic observations for decadal forecasting, the same rationale should hold for the long-term models. At this juncture there are only two CMIP5 groups that can rest on their laurels, the rest have to re-think (that would include UKMO) and come up new configurations. Even the two apparently valid groups will have to stand the test of time over the next 5 yrs as I said earlier.

The UKMO says this:-

“The latest decadal forecast, issued in December 2012, show that the Earth is expected to maintain the record warmth that has been observed over the last decade,”

That’s for 5 yrs and I think that’s reasonable but with a solar recession upon us that wont last forever.

“….and furthermore a substantial proportion of the forecasts show that new record global temperatures may be reached in the next 5 years.”

As I asked up-thread, why? By what mechanism? There is no single mechanism evident in the spaghetti. This seems to imply that the 5 yrs will be dominated by El Nino’s but this is what they say about that:-

“….we know that the predictability of phenomena like El Nino/La Nina is limited to at most a year in advance.”

They contradict “new record global temperatures” with:-

“Similarly relative cooling in the north-east and south-east Pacific Ocean compared to other parts of the global oceans is also evident in Figure 3. The pattern of this cooling is similar to that observed in the Pacific Decadal Oscillation, and has been a notable difference in ocean temperatures between the 1990s and 2000s (Figure 4). Were this cold phase to continue as forecast, this would act to moderate global warming in the next few years, as it has over the last decade.”

So where does the record level arise? Oh yes:-

“Decadal forecasting is immensely valuable, scientifically, because it represents a stringent test of how well the model simulates natural variability and also how well it captures the longer term anthropogenic warming trend”

But what if that “longer term anthropogenic warming trend” that they assume, is non-existent because it is based on a fallacy as below?

http://tallbloke.files.wordpress.com/2010/07/eggert-co2.png

We’ll see over the next 5 yrs for sure.

The white lines show hindcasts, ie: model simulations started from older initial conditions and then run onwards, and compared with the observations to see how well the model does. The point here is that the hindcasts with the new model (HadGEM3) compare better with the observations than the old model (HadCM3) and so this gives us more confidence in the new model.

These decadal forecasts use “initialised forecasting” techniques, ie: the models are started at the observed state for the current time – as distinct from the long-term climate projections that start back in pre-industrial times, run through the 20th Century and then on into the 21st Century, meaning that they can’t be expected to capture the exact year-by-year variations that the initialised forecasts are attempting to capture. Because the initialised forecasts are started off at, say, the right place in an ENSO cycle, they potentially can capture the natural variations arising from ENSO and other modes of variability. This is still early days of course, there is still a lot more work to do, but you can see from the 2012 figure that the hindcasts show the model agreeing with the observations reasonably well (and better than the HadCM3 hindcasts as shown in the 2011 figure).

The first time that these initialised forecasting techniques were used for decadal forecasting was this paper published in 2007. So this was the first time there was actually a proper forecast looking forward in time – anything before then is a hindcast. This is the case for all versions of the decadal forecast that you might find.

http://www.bishop-hill.net/blog/2013/1/10/spot-the-difference.html

The thread at Bishop Hill contains a lot of useful info

http://www.bishop-hill.net/blog/2013/1/12/lewis-on-schmidt-on-climate-sensitivity.html

The discussion centres around Nic Lewis’s recalculations of climate sensitivity (S) which was subsequently reported in the WSJ by Matt Ridley.
Lewis’s later work focused on recalculations of S using updated figures for aerosol forcing. This brought S down to a value of 1.75 degC

The other recent work by Nic Lewis was the discussion at Judith Curry’s about the use of uniform priors on Forster and Gregory 06 (I’ll say a bit more about Bayes and priors at the end of this comment for the newcomer)

http://judithcurry.com/2011/07/05/the-ipccs-alteration-of-forster-gregorys-model-independent-climate-sensitivity-results/

FG2006 was a paper based on empirical observations alone, unlike other IPCC studies that use models to some degree or other.

The original graph showed a tightly constrained probability density function (PDF) for S, with a modal (average) value of around 1 degree, which you can see in Fig 1 on the Curry page.

The IPCC then decided to modify the PDF by applying a uniform prior on S. What this means is that they made a claim (a prior assumption – or prior in Bayesian terms), that S was constrained to lie between 0 and 18.5 degrees, with an equal probability of any of those values being correct.
By then applies Bayes Rule, the IPCC found a posterior PDF ( the one that comes out of Bayes by applying the prior to your evidence – the actual climate data) that was much more flattened out, with a much fatter tail. This also had the effect to push up the modal value of S (at the peak of the graph), and the median too.

http://curryja.files.wordpress.com/2011/07/fig4_influence-of-prior.jpg

There are a number of issues with this approach, and Nic Lewis explains a lot better than I can. One is that FG2006 was not a Bayesian study, so using priors may not be appropriate. Secondly, chosing a uniform prior on S may not be appropriate. As Lewis point out, the relationship between S and forcing Y is inverse, so a uniform prior in Y would give a completely different result.

Where Dave Frame comes into the picture is his 2005 paper
“Constraining climate forecasts: The role of prior assumptions”
http://www.climateprediction.net/science/pubs/2004GL022241.pdf

which seems to get cited a lot in the literature.

Nic Lewis, in response to a comment by me at BH, writes thus:

Andy Scrase
“Frame et al seems a well cited paper on priors in climate prediction”

Or badly cited, depending on one’s viewpoint! It is the only place I have ever read the proposal that the prior used for estimating climate sensitivity (or indeed any parameter in any field) should depend on the purpose to which the estimate is to be put. That seems to me to go against both probability theory and scientific principles.

I can understand multiplying a probabilistic parameter estimate (a posterior PDF, in a Bayesian context) different loss functions, according to the use of the estimate, but not changing the estimate itself – which is what Frame et al (2005) advocate, through varying the prior used.

I would be interested in the views on this point of any scientists and/or statisticians who read this.

So this is where I am up to – trying to figure out the literature and reading some textbooks too.

As a postscript, for anyone not familiar with Bayes, then the book recommended to me at BH is a great historical introduction to the subject

“The Theory That Would Not Die: How Bayes’ Rule Cracked the Enigma Code, Hunted Down Russian Submarines, and Emerged Triumphant from Two Centuries of Controversy”
http://www.amazon.com/dp/0300188226

In general terms, Bayes is a simple rule that uses prior assumptions with observed facts to create an improved model or PDF, the so-called posterior
Bayes gets used in things like anti-spam filters, where we can start with a uniform prior that all mail is not spam, and then “train” the model by flagging spam. Bayes then works to find a PDF that a certain email is spam. This then gets used iteratively so that the posterior gets used as the new prior.

BTW, occurred to me that NIWA’s regional climate model is derived from UKMO’s Unified Model (UM). Does HadGEM3 make that obsolete I wonder?