Saturday, January 14, 2012

The Whitehouse bet

I hear that David Whitehouse is crowing about winning the bet we had over global temperatures. I can't say I blame him - but he had better make the most of it, as the nature of the ongoing warming trend is such that it will not be long before the 1998 temperature estimate is unambiguously beaten in all three of the major temperature analyses. It already has been in two of the data sets, of course, but we did agree to use the Hadley centre analysis. (UPDATE but thanks to commenter doskonaleszare have a look at this!)

More or Less covered this yesterday (you can listen on line, or get the podcast), as they did with the original bet. Tellingly, you can hear that David Whitehouse was not prepared to bet against a record over the next 4 years on the same basis. So even though he continues to bluster about a standstill in temperature, he obviously doesn't really really believe it. Here are the three data sets (offset for clarity), with the simple linear trend (dotted lines with associated decadal trend value) over the last 20 years. I estimated observational values for 2011, as they are not actually published yet. It is is quite clear that even with the supposed halt in warming, the trend continues to be positive, and the 1998 value (dashed line) will be regularly exceeded in a few years by all measures.

That said, there is little sign of the acceleration in warming that most models had predicted, and it increasingly seems that the Smith et al forecast (for example) was a bit excessive. This new paper also suggests that the transient response of a modern model (albeit a particularly sensitive one) has to be significantly downscaled to match observations. Mind you, that paper also has a worrying discrepancy between the results obtained with 1900-2000, versus 1850-2010 data. Normally one would expect the latter to be broadly a subset of the former - more data means closer convergence to the true value - but the two sets of results are virtually disjoint, which suggests something a bit strange may be going on in the analysis (cf Schmitter et al with the land-only versus land+ocean results). But just a glance at the first figure shows a striking divergence between model and data over the first decade of the 21st century (compared to the close agreement prior to then). Something isn't quite right there.


doskonaleszare said...

My 2 eurocents:
1. 1998 is no longer a record year in the HadCRUT4 dataset, thanks to better Arctic coverage. But alas it is not published yet.
2. It's not only CanESM2. For some reason, many of the CMIP5 models significantly overpredict warming for the last decade.

James Annan said...

Thanks, have you any more details on HadCRUT4?

doskonaleszare said...

According to Dr Chris Gordon, who gave a talk at the recent MOSAC meeting,

"The new version of the global near-surface temperature anomaly dataset, HadCRUT4 is close to completion. The dataset is produced in a collaboration between the Met Office Hadley Centre and the Climatic Research Unit of the University of East Anglia. The new analysis includes new data in regions that were poorly observed in HadCRUT3, particularly in Russia and at high latitudes in the northern hemisphere. The underlying data used to generate the new HadCRUT4 dataset will be free from any restrictions to public distribution when the dataset is released.
The manner in which uncertainties are represented in the dataset has been updated to allow more straightforward analysis of the data by scientists. The dataset is now presented as an ensemble dataset, with uncertainties communicated using multiple realisations of the data that are equally likely given the uncertainty model. This ensemble approach has not previously been applied to global surface temperature datasets.
The update to the data set has adjusted the annual temperature anomalies in comparison to the previous version of the dataset. Most notably, the inclusion of new bias adjustments for marine data has resulted in a warming in the mid 20th century, relative to HadCRUT3. The inclusion of new land station data at high latitudes and in Russia has resulted in a warming of years in the late 20th century/early 21st century. These temperature differences, although small, make some differences to the annual temperature rankings but long-term trends in HadCRUT4 are not greatly different than those of HadCRUT3.
Once accepted by peer review, the HadCRUT4 dataset will supersede HadCRUT3 as the principal dataset used by the Met Office Hadley Centre to assess global temperature change."
(slide #6)

James Annan said...


dana1981 said...

Dr. Annan, we're going to publish a post regarding this wager on Skeptical Science on Tuesday, and if you have the time, I was hoping you could make sure we've got all the facts right.

Carrick said...

How much warming we see in the next 30 years also depends on how much pollution is generated in the developing 3rd world.

You've probably seen this already. Note there could be no observable warming for as long as 30 years, depending on when pollution controls kick on.

I tend to favor the "late pollution controls", because (IMO) that is more realistic of human nature.

crandles said...


"In fact in 2011, La Niña had the fifth-strongest cooling effect on any year since 1950, and nevertheless was the hottest La Niña year on record, according to the World Meteorological Association. 2009 and 2010 both saw relatively moderate ENSO conditions, and thus were Annan's only real chances of winning the wager."

While I wouldn't really say this is wrong, it would be better if you account for the lag between ENSO and its effect on global average temperature. If you account for a 3 month lag or a 1 to 6 month lag then 2008, 2009 and 2011 didn't really offer much hope for Dr Annan but 2010 was an excellent chance.

dana1981 said...

crandles - I actually did account for a 3 month lag. The average MEI then influencing 2009 was slightly negative, and 2010 was slightly positive, but both were relatively small.

Though actually with the news of HadCRUT4, I'm going to re-work the post a bit.

Jesús R. said...

"there is little sign of the acceleration in warming that most models had predicted,"

So what's your take on that?

a) At RC, the usual suspect used to be variability (weather) masking the underlying trend.

b) Now, the Gillet et al paper suggests a lower transient response to climate forcing.

c) Skeptics, however, would love the lower climate sensitivity option.

d) Some papers even pointed to a lower-than-modelled radiative forcing (Kaufmann et al or Solomon et al, if I didn't misread them).

What a mess :)

James Annan said...

I would say, a bit of a mix of several things...

Slightly low solar minimum

A bit of variability (eg Tsonis et al work)

Slightly lower transient sensitivity, which then points to (probably) slightly lower equilibrium sensitivity too.

I haven't yet checked on doskonaleszare's statement about the other IPCC models - Gillett do indicate that their model is probably one of the worst, but it seems that the NCAR model is pretty bad too. For the AR4 (eg Knappenberger analysis), the discrepancy was perhaps close to visible but not yet decisive IMO.

doskonaleszare said...

more examples here

I wonder if a reduced radiative forcing (e.g. Skeie, R. B., Berntsen, T. K., Myhre, G., Tanaka, K., Kvalevåg, M. M., and Hoyle, C. R.: Anthropogenic radiative forcing time series from pre-industrial times until 2010, Atmos. Chem. Phys., 11, 11827-11857, doi:10.5194/acp-11-11827-2011) could explain this discrepancy. AFAIR Meehl et al. did some sensitivity tests for the CCSM4, but only for the direct effect of sulfate aerosols, and concluded that simply increasing the negative forcing would not work that well.

Jesús R. said...


I'm surprised James mentioned Tsonis for variability, because

1) I always prefer more physic-based papers such as Foster & Rahmstorf removing ENSO and so on. Statistical analysis always sound to me like curve-fitting and reminds me of thinks like correlation is not causation. :)

2) Swanson (co-author with Tsonis) argued at RC that the higher variability they found, implied a somewhat higher sensitivity:

3) Anecdotally, Tsonis & Swanson found an underlying accelerating warming trend (though I don't know if similar to those shown by the above-mentioned models):

Finally, I understand that it's not clear whether the observed trend is below model projections, but I'm always surprised that Tamino and RC are usually so sure that it isn't. Sometimes it looks to me as if they were saying "this is what models projected" rather than "we don't know yet whether this is what models projected".


James Annan said...

doskonaleszare, thanks that looks pretty bad. I would guess that these models will actually fail standard D&A tests (as the anthrop coefficient will be strictly less than 1), which will cause some raised eyebrows.

Jesús, I was interested in the way that the Tsonis thing tended to make curves into linear segments. OTOH Tamino is often banging on about how easy it is to fool yourself that linear segments appear when they don't really exist. It's nothing more than an interesting idea to me.

Knappenberger et al looks pretty sound to me, though one can debate the precise language :-)

James Annan said...

BTW, do you have an equivalent plot for CNRM-C5? Gillett et al find a GHG scaling coefficient centred on 1 for this model (their supplementary), and it seems to have an overall transient response rather lower than observed over the 20th century (eg fig 18 here). However it has a mid-range sensitivity, therefore presumably a very large ocean heat uptake (unless the forcings are strange).

doskonaleszare said...

Bah, the KNMI doesn't have a complete RCP45 ensemble for CNRM-C5.


other "good" CMIP5 models

James Annan said...

Ah, I was assuming that your previous figure was fairly representative rather than specifically the worst ones :-)

doskonaleszare said...

Indeed I eyeballed the worst ones, sorry for the confusion.

Nevertheless, the CMIP5 model spread will be larger than the CMIP3, and the interpretation of multi-model ensembles will become tricky (should we weight the models according to their 20CEN skill? bias-correct them, as described in Gillett et al?). I bet the relevant chapters of AR5 FOD make an interesting read ;)

Paul S said...

But just a glance at the first figure shows a striking divergence between model and data over the first decade of the 21st century (compared to the close agreement prior to then).

I had this discussion with Judith Curry regarding CCSM3. I think the appearance of close agreement in the latter part of the 20th Century is a mirage caused by the two volcanic events, in 1982 and 1991, plus the large 1998 El Nino.

Looking at Gillett et al's Figure 1a, people will tend to look through the data to determine the trend. However, the variability between 1982 and 1997 is clearly dominated by volcanic events. If you want to determine the continuous and continuing trend in the data you need to draw lines across the top of the pits rather than through them. If you do that it's pretty easy to see a linear trend slowly diverging from observations since about 1970. The only reason this divergence has been uncovered in the past decade is the lack of a large volcanic eruption.

Jesús R. said...

doskonaleszare, could you please give me some clue to find the Meehl et al paper you mentioned here?:

"AFAIR Meehl et al. did some sensitivity tests for the CCSM4, but only for the direct effect of sulfate aerosols, and concluded that simply increasing the negative forcing would not work that well."

doskonaleszare said...

Climate system response to external forcings and climate change projections in CCSM4 from the CCSM4 Special Collection JClim

free pdf here

Jesús R. said...

Thanks, doskonaleszare! :)