Sunday, February 22, 2009

Decadal prediction cage fight

We have just had the traditional "Let's spend our spare money before the end of the financial year by inviting a few foreigners to visit" February workshop again. Last year I helped to organise the equivalent event, but this year I didn't have any spare money so it was up to others. As a result, it was rather boring :-) but there was an interesting session about decadal prediction, including contrasting presentations from Noel Keenlyside (of decadal cooling fame) and Nick Dunstone who is a recent addition to the Hadley Centre group. First, Keenlyside presented an extended version of the work underpinning his heavily-criticised Nature paper. He started off by claiming that there was significant internal variability in the global temperature signal which was responsible for perhaps as much as 30% of the recent (30y) warming. This was justified by a simple calculation in which he started with the global temperature series and subtracted a scaled and 11y-lagged CO2 forcing curve! I found this deeply unimpressive, as it is well known that there are other significant forcing factors which can explain a lot of the wiggles, leaving a much smaller amount of natural variability (and arguably the natural component could even have been a cooling):


(From Global Warming Art via Wikipedia)

To be fair he had other modelling results suggesting a link between variability in the Atlantic meridional overturning circulation (AMOC) and global temperature variability, but given how dodgy the first part (which I am more familiar with) was, I'm reluctant to take the rest of it on trust. He then finished with results from his model initialisation and prediction system, which as RC and others have already noted: (a) is physically dubious (b) has negative skill (compared to the unitialised model) (c) already gave two false alarms of cooling in their hindcasts over the past 50 years and (d) has clearly gone badly wrong at the end of the 20th century. I didn't press him too hard on the issue of the bet (offered by RC) that he has apparently refused, but he did concede that his future prediction was probably a bit on the cold side.

Nick Dunstone then presented a fairly comprehensive set of idealised identical twin tests to evaluate the importance of different observational (surface versus deep ocean) data sets for multi-year prediction ("decadal" seems optimistic really, as there isn't really much skill at 10y for any system). One of the tests they tried was the simple sea surface temperature nudging scheme that Keenlyside used, and they found this gave poor results including a very unrealistic AMOC. Their own scheme which uses deep ocean data gave clearly superior results. I wouldn't claim the HC work is the last word on the matter but it was far more convincing to me. Their latest updated prediction is much the same as the one in their Science paper, just starting off marginally cooler due to the 2007 La Nina, but still expecting to see new record warm years popping up pretty soon. Good news for my bet, although I think they would put the odds a little lower than Hansen :-)

The Japanese group still seem to be developing about 3 or 4 different data assimilation schemes for their decadal predictions. In the closing discussion the leader of this project openly admitted - indeed, it almost sounded like a boast - that he had no idea if the predictions would be of any real value to anyone even if they do show some skill, he was just doing it because it was scientifically interesting. I wonder if he puts it quite that plainly when speaking to funding agencies :-)

12 comments:

Hank Roberts said...

James (realizing this question might also be addressed directly to globalwarmingart) -- on that illustration, would it be reasonable to take the difference between "observed" and the area outside the "predicted" gray-fuzz range and chart that as "missing/unexplained" (or somehow express how much of a gap remains)?

James Annan said...

Basically yes, that gap indicates a gap in our understanding (as a lower bound!), although note that we would expect a little bit of mismatch at the 95% level anyway (not as much as appears in the plot though).

As a possibly non-exhaustive and arbitrarily-ordered list of reasons for the discrepancies:

The external forcings are imperfect, and some marginally significant ones are missing (eg land use). The obs have some biases (remember this, which I expect to slightly improve the fit once a correction is made) and even if these were all corrected, the global time series is not precise anyway due to limited coverage (note how gistemp and hadcru differ even though they use almost precisely the same original obs data). The model may have inadequate natural variability on the appropriate time scales - I don't know about this model in particular, but none of them really do ENSO/PDO/NAO "correctly". And last but perhaps not least, the model response to the external forcings is certainly not perfect anyway.

But we can clearly do a far better job with a comprehensive analysis than by just using CO2 forcing with an 11 year lag!

John Finn said...

James

Looking at the graphic it would seem that, according to the model, solar + ghg are the major factors which contribute to the early 20th century warming and that the increase in sulphate (industrial)aerosols is the factor which not only halted the warming trend but actually reversed it.

I have to say I have some doubts about this, industrial aerosols, unlike CO2, are short-lived in the atmosphere and do not build up over time. There would need to be a massive and constant increase to produce the claimed effect. But that's not all.

Because of the short residence time, the climatic effect of aerosols will be most pronounced over the regions from which the aerosols originate. Mike Mann and Phil Jones published a paper in ~2003 which stated that the effect of aerosols is "regionally specific". So, the areas that should have experienced the greatest cooling in te post-1940 period are the industrialised regions in the US and W. Europe. But that's not what happpened.

The region which experienced the largest (by far) drop in temperatures was the ... the Arctic.

According to the GISS zonal data record, the Arctic (64N-90N) cooled almost 1 deg C or about 4 times as much as the indusrialised mid-latitiude regions - and about 3times the rate of my home city in the UK midlands. I'm not sure if you're originally from the UK, James, but if you are you'll know that during the 1950s, 60s and 70s, the midlands was one of the most intensely industrialised regions in the world.

This GISS Arctic data provides other points of interest such as the fact that the arctic warmed almost 2 deg C during the 1910-1940period, i.e. far more than any other region. I don't accept that this was due to ghgs (or the sun). Even in 1958 the atmospheric CO2 concentration was only ~315 ppm - equivalent to an increased forcing of ~0.5 w/m2 above pre-industrial levels.

The reconstruction model is missing a huge factor which is masked by the averaging of temperatures over the world.

My bet that missing factor is the ocean.

Hank Roberts said...

Sorry, James, I mentioned the thread at Tamino's yesterday. My bad.

James Annan said...

no problem Hank - but it does explain things :-)

John,

I don't think it is necessary to explain every wiggle in terms of deterministic factors - there is undoubtedly some natural variability in there too (in increasing amounts as we look at smaller areas). Eg, I believe there was a big El Nino around the 1940 peak. But there's a big volcano in 62-63, and a small drop in solar forcing after ~1945 too, to add to the aerosol forcing.

Of course all the models include the ocean!

Steve Bloom said...

It's clear from this new paper that teasing out the specific effect of aerosols isn't entirely simple even when good obs are available. Re the Arctic, IIRC some work has been done pinning at least part of the early 20th-century warming to industrial soot.

John Finn said...

I don't think it is necessary to explain every wiggle in terms of deterministic factors - there is undoubtedly some natural variability in there too (in increasing amounts as we look at smaller areas) Eg, I believe there was a big El Nino around the 1940 peak. But there's a big volcano in 62-63, and a small drop in solar forcing after ~1945 too, to add to the aerosol forcing

No, James, that does not explain why the Arctic warmed 2 deg between 1910 and 1940 and cooled a degree between 1940 and 1970. There are not wiggles they are sustained long term climate shifts which are, far and away, the biggest contributors to climate change over the past century.

Of course all the models include the ocean!

Ok. Yes, James, I know all the models "include the ocean". It's a bit disappointing you've reverted to the usual tactic (on both sides) of deliberately misinterpreting my point.

You graphic is effectively the result of a "detection and attribution" exercise. It attempts to explain warming (and cooling) over the past century by various forcings. But it doesn't make sense.

QUESTION: If industrial aerosols are responsible for the mid 20th century cooling - why is it that the Arctic cooled at 4 times the rate of the NH in general.

Perhaps Hank can answer the question .... I can wait

Adder said...

John,

The transport of aerosols in the atmosphere is a complex process - if they get into the free troposphere, their lifetime is 1-2 weeks, and during that time, they can travel long distances, such as China to Europe, or Africa to America. Sure, their climatic effects are mostly "regionally specific", but there are exceptions. Google "arctic haze".

John Finn said...

The transport of aerosols in the atmosphere is a complex process - if they get into the free troposphere, their lifetime is 1-2 weeks, and during that time, they can travel long distances, such as China to Europe, or Africa to America. Sure, their climatic effects are mostly "regionally specific", but there are exceptions. Google "arctic haze".

Totally agree with all you've written. But te net effect of "arctic haze" is one of warming . My point still stands, i.e. aerosols cannot explain the 1940-70 arctic cooling and CO2 (or solar) cannot
explain the 1910-1940 arctic warming.

Hank Roberts said...

Asked and answered elsewhere (dozens of places). One was
http://tamino.wordpress.com/2008/11/23/open-thread-8/#comment-24387

Steve Bloom said...

James, I assume that by now you've seen this article. If the translation is accurate, your concerns about management would seem to be confirmed. Any details you could provide on this report would be appreciated.

bi -- International Journal of Inactivism said...

Steve Bloom, James Annan:

A link to the complete report in the original Japanese.