Tuesday, July 05, 2005

10C? Not likely!

I've found myself pointing people in the direction of this RealClimate page a few times, so I thought I might as well post my own thoughts on the possibility of extremely high climate sensitivity (= equilibrium temperature change for doubled atmospheric CO2 concentration) that has been suggested in several papers, and most recently Andreae, Jones and Cox (Nature 2005, subsequently AJC).

The basic approach that AJC and several previous papers use is to look at the following globally-integrated heat balance equation

c dT/dt = DQ - L.DT ... (1)

which describes how the average temperature (T) changes through time in response to radiative forcing DQ, with c being the climate system's heat capacity and L describing the radiative response to temperature change ('d' indicates a derivative, 'D' is a finite perturbation from the pre-industrial equilibrium). L determines climate sensitivity via

DT2 = DQ2/L ... (2)

where DQ2 is the radiative forcing due to doubled CO2 (about 3.7 W/m2). Note that eqn 2 is just the steady state version of eqn 1, rearranged and evaluated at doubled CO2.

Eqn 1 rearranges to

L = (DQ - c dT/dt) / DT ... (3)

We have estimates for all of the terms on the RHS of eqn 3 over the past 40-100 years or more, and so in principle we can just plug them in, out drops L and thus climate sensitivity.

Unfortunately, the terms on the RHS of eqn 3 are not known precisely, but all have uncertainties associated with them. In particular, the net radiative forcing DQ is highly uncertain. The radiative forcing has two major components, anthropogenic CO2 and sulphate aerosols. Although the radiative forcing of CO2 is well-known to be substantial and positive (currently about 2.4 W/m2), the aerosol effect is highly uncertain, with estimates ranging from zero to a negative forcing with greater magnitude than the CO2 effect. When Gregory et al did this calculation back in 2001, they found that L could turn out to be zero or even negative, meaning an unbounded or (absurdly) negative climate sensitivity. We have two major unknowns - DQ and L - and therefore (with hindsight) it is hardly surprising that we cannot simultaneously constrain them with a single equation.

The AJC paper is already being publicised as "new research" that "global warming looks set to be much worse than previously forecast" (eg in this NewScientist article). It is, of course, no such thing. It is a "Progress" article, in other words a review paper which by design does not present substantially new research. The authors do include a simple carbon cycle feedback, which makes the warming slightly worse, but other than that they basically repeat an old approach which is well-known to not provide a useful constraint on climate sensitivity.

Now, I certainly don't mean to be critical of those who originally tried this approach and found it unhelpful, but having got the result, and understood the reasons behind it, it would seem more productive to look for other ways of estimating climate sensitivity rather than repeat this method that cannot work and throw ones hands in the air in despair about how dreadfully scary everything is and how WE'RE ALL GOING TO DIE UNLESS WE STOP BURNING FOSSIL FUELS.

This heat balance approach in no way indicates that the more moderate range of values for climate sensitivity previously presented by the IPCC (1.5-4.5C) is any less likely than was previously thought - it only shows that this one new approach does not provide a useful constraint. Fortunately, there are many other lines of argument that can be used to constrain climate sensitivity (including evidence from paleoclimate, post-volcanic cooling, and complex GCM modelling), and although these approaches also have potential weaknesses and limitations, they all basically agree that the most plausible value for climate sensitivity is around the 3-4C mark, and some of them also seem to clearly rule out anything approaching 10C. In this context, the claim in the accompanying News article seems more than a little exaggerated:
Andreae acknowledges that there are many uncertainties about his study. But he points out that it is the best estimate we have so far. "This forces us to accept that pessimistic climate scenarios are much more plausible than had been thought," he says.
Sorry, but I do not accept this at all. There are plenty of much better estimates. There may be a case for raising the IPCC range marginally (say to 2-5C, or 1.5-6C if they want a firmer confidence interval than the previous "likely") but I would be very very surprised to see it go any higher than that.

Update: Gavin Schmidt gives a more detailed analysis of the AJC paper on RealClimate here, and draws much the same conclusion.


Anonymous said...

I doubt this is the sort of bet you would be looking for against 'alarmists'. However have you seen the dichotomy of opinion shown on p41 of this Allen & Frame presentation? Does this offer any prospect of a bet over which way that will be resolved?

That presentation is obviously partly based on Constraining climate forecasts: The role of prior assumptions.

Does this make your comment "I would be very very surprised to see it go any higher than that" a bit too dismissive of higher sensitivities?

I agree that high sensitivities are not likely, just want to hear some views from the other side of that dichotomy. I don't want to just swallow what I am hearing from climateprediction.net when the conclusions seem so different.

Hope you don't mind me putting a comment on RC then adding more here.

Chris Randles

James Annan said...

On that slide, a lot depends on what "negligible" means, and anyway it will be very hard to determine climate sensitivity accurately.

I'm not sure there is that huge a gulf in opinions really. It is partly a matter of perspectives - they would be right to say that the chance of high sensitivity cannot be absolutely categorically ruled out. I am right to say it is small.

The question is, how small - and it is not easy to assign probabilistic estimates to things like this - especially towards the tail of the distribution (everyone agrees that about 3C is most likely, give or take a bit - even the sceptics say ~1.5C, and the highest "best estimate" I have seen is about 4.5C). Assessing the tails of the distribution requires a lot of judgement and not just a better formula.

One issue may be that each researcher believes his own method is the "best" and gives the "right" answer. However, the correct answer for climate sensitivity must fall in the intersection of the different estimates, not their union. An estimate that relies on a (small) subset of the available evidence is likely to be too uncertain. On the other hand, if we underestimate various uncertainties that go in to these "optimal"techniques (which is always a risk) then we get an estimate that is unrealistically constrained (and therefore may not include the true answer).

My bottom line is that no-one has yet produced a model with high climate sensitivity (>6C, say) which gives a plausible description of the Earth's climate, including eg a hindcast of the last century and plausible paleoclimate simulations. About 5-6C seems like a upper limit to me.

I will write a longer post about probabilistic prediction soon.

Anonymous said...

Strikes me that not all of the parameter uncertainties are uncorrelated.

georgesdelatour said...

James, what do you make of this?


James Annan said...


I talked about that here and briefly again here. In summary, there is enough that is dodgy that I'm not prepared to trust it, although I do not know how significant the problems with it are...but for now I certainly prefer the IPCC!