Thursday, April 07, 2016

Climate sensitivity is 5.3C?

Too late for an April Fool, my eye was caught by this headline in the Guardian:
 With the article itself containing the odd phrasing:
Researchers said that a doubling of carbon dioxide in the Earth’s
atmosphere compared with pre-industrial times could result in a global
temperature increase of up to 5.3C – far warmer than the 4.6C older
models predict.
4.6C seems a strangely high value to start with, it's some way from any consensus value (and it's not even the upper end of the oft-quoted undcertainty range of 2-4.5C).

Sadly, the paper is paywalled, so I can only see the abstract. But it seems that the authors are arguing that models have a bias in the way they represent clouds (too icy, not wet enough), and correcting this bias will shift their sensitivities up a bit. 4.6C is the most sensitive model in the AR4 and it seems to get shifted up a smidgin to 5.0-5.3C (there is presumably some uncertainty in the adjustment to each model). The paper also says the sensitivities shift by "up to 1.3C" which must therefore be the largest potential shift of any of the models, and probably some way above the mean estimated change. It doesn't look like such a revolutionary change as to justify the headline, even if one accepts that the result is right.

Of course, if the models were more sensitive, it would (other things being equal) result in greater warming not only in the future, but also in the past. Which would make it more challenging to reconcile with what we've observed. Kevin Trenberth (who has presuably seen the whole paper) doesn't seem that impressed. Maybe I'll manage to find a copy somehow...

Update: seems like several others have weighed in similarly (Gavin, ATTP). The Yale puff piece is particularly misleading, constrasting the old 2-4.6 model range with a new range of 5-5.3 which is surely nonsensical. Even without reading the paper, it's obvious enough that 5-5.3 does not represent a new range that would contain all the models, since they only move by "up to 1.3C". I'm fairly confident that my interpretation above must be correct.

I don't want it to sound like I'm determined to reject any attempts to change our estimate of sensitivity. But any substantial changes will have to be supported by significant evidence to overturn the weight already accumulated. In this case, what I'm really objecting to is the hyped-up presentation of what is actually a pretty small change anyway.


Anonymous said...

Gavin's been having a bit of a Twitter rant about it. The press release seems a model example of how not to promote your paper. I agree that 5.3C for ECS seems bizarrely high and very hard to reconcile with all the other information.

Dan said...

Most annoying sentence from the paper's conclusion (IMO): "Should the low-SLF bias be eliminated in GCMs, the most likely range of ECS should shift to higher values."

EliRabett said...

So Hansen et al 1988 was right after all:)

Hank Roberts said...

UPI especially screws up (at least I hope it was UPI and not the original press release) saying the paper is published in Nature, and providing a clickable "Nature" link that goes to Science instead.

In a new study published in the journal Nature, researchers ...

steven said...


is your friend

James Annan said...

Great resource, thanks!

Seems that the paper is even worse than I had thought possible. It's just a perturbed parameter ensemble on CAM (atmosphere only!) and the 5 and 5.3C values are the two "best" simulations according to their particular metric on cloud water.

If they believe these models are really better overall than the control, hopefully they'll do the full raft of CMIP simulations so their performance can be robustly evaluated. I'm not holding my breath.