Friday, May 30, 2014

Can we trust climate models?

Our latest paper, Can we trust climate models? has just appeared on the Wiley website (open access thanks to our Japanese friends who paid for this). To save a click, here's the abstract:
What are the predictions of climate models, should we believe them, and are they falsifiable? Probably the most iconic and influential result arising from climate models is the prediction that, dependent on the rate of increase of CO2 emissions, global and annual mean temperature will rise by around 2–4°C over the 21st century. We argue that this result is indeed credible, as are the supplementary predictions that the land will on average warm by around 50% more than the oceans, high latitudes more than the tropics, and that the hydrological cycle will generally intensify. Beyond these and similar broad statements, however, we presently find little evidence of trustworthy predictions at fine spatial scale and annual to decadal timescale from climate models.
The paper was invited by the editors some time ago, under a slightly different title but they didn't object to us changing this to something that we felt matched the content a little better. The original plan was for two papers to appear together, with the other one written by a prominent sceptic. However, this seems to have fallen by the wayside. Although we could probably have guessed what they were likely to write, we didn't really want to take part in a direct debate, so rather than aguing against straw man criticisms we just tried to set out our own ideas. Given how it turned out I'm glad we did this! Writing this paper also fitted in well to our personal plans, as we had already pretty much decided at that time to leave Japan, and this gave us a good opportunity to summarise and review some work (including our own) when we weren't really minded to embark on a big new ambitious plan of research. I don't think our argument is likely to be considered controversial - perhaps some might think we are a little pessimistic about the regional performance of models (by which we basically mean anything less than hemispheric) but our recent work with paleoclimate really has brought home to us that they don't get much right about patterns on this sort of scale, even for temperature still less precipitation.

The reviewers made some helpful comments, perhaps the only real criticism was that we sounded like a bland consensus and weren't really making an opinionated statement as might be hoped for in an "opinion" article. But we wanted to say what we thought was correct and justified (ie, a true summary of our opinions), rather than being artificially controversial. I'll be very happy if the paper is seen as a useful summary of how much we can trust climate models, and why.


Nick Stokes said...

"The reviewers made some helpful comments, perhaps the only real criticism was that we sounded like a bland consensus"

Yes, that was my feeling too. The paper tends to build the expectation that someone is making predictions with the GCM's that are wrong. But it never gets to that. It basically says we can do seasonal scale, and we can do long term. Who disputes that?

It seems to me that GCM's generate random synthetic weather that responds to forcing. As such, you can average to deduce climate responses. They don't claim that the random weather is what will happen, unless they have been properly initialised, in which case seasonal (from that time) is about the time scale.

crandles said...

"The large-scale understanding of the physics seems to be sufficient, but the details are either not well understood, or are not being sufficiently well approximated by the model code."

Bland consensus or refusing to stick your neck out regarding what is most important?

Hank Roberts said...

Details, details. Do global climate models have a way to run scenarios like this, in which a sea level rise less than one meter uncorks a very large area of ice?

Earlier, this (and the reports that meltwater ponds disappeared from ice sheets before they broke up) seemed to start giving details that suggest rapid change:

"Here we present a numerical model that simulates the disparate calving regimes observed, including the detachment of large tabular bergs from floating ice tongues, the disintegration of ice shelves and the capsizing of smaller bergs from grounded glaciers that terminate in deep water. Our model treats glacier ice as a granular material made of interacting boulders of ice that are bonded together. Simulations suggest that different calving regimes are controlled by glacier geometry, which controls the stress state within the glacier. We also find that calving is a two-stage process that requires both ice fracture and transport of detached icebergs away from the calving front. We suggest that, as a result, rapid iceberg discharge is possible in regions where highly crevassed glaciers are grounded deep beneath sea level, indicating portions of Greenland and Antarctica that may be vulnerable to rapid ice loss through catastrophic disintegration."

Viewing that in the light of this very new one:

"“East Antarctica’s Wilkes Basin is like a bottle on a slant,” says lead-author Matthias Mengel, “once uncorked, it empties out.” The basin is the largest region of marine ice on rocky ground in East Antarctica. Currently a rim of ice at the coast holds the ice behind in place: like a cork holding back the content of a bottle. While the air over Antarctica remains cold, warming oceans can cause ice loss on the coast. Ice melting could make this relatively small cork disappear – once lost, this would trigger a long term sea-level rise of 300-400 centimeters. “The full sea-level rise would ultimately be up to 80 times bigger than the initial melting of the ice cork,” says co-author Anders Levermann.

“Until recently, only West Antarctica was considered unstable, but now we know that its ten times bigger counterpart in the East might also be at risk,” says Levermann, who is head of PIK’s research area Global Adaptation Strategies and a lead-author of the sea-level change chapter of the most recent scientific assessment report by the Intergovernmental Panel on Climate Change, IPCC...."

I see some concern, e.g.:

"... Models from Mengel & Levermann (2014) also show that "We have probably overestimated the stability of East Antarctica so far". [Remark: At EGU2014 Vermeersen and Pollard gave presentations showing WAIS collapse in a few centuries, and a large EAIS response on longer time scales. I assume that this work will be out really soon, so i will update this post when I know more.] ..."

I suppose it's hard to perturb a global model with something like that.
But they do handle throwing in the occasional volcanic eruption.
What about throwing in an abrupt sea level change?

Hank Roberts said...

blush. Sorry, you were on it already, just found this:

jules said...


Have you read the paper, or are you judging the whole things from a single quote? I thought we were clear about the embarrasssment of model failures for regional and decadal prediction. With failures that profound, there's not much point going into little details. Not really sure what else there is to say.


crandles said...

I did read the whole paper and you were clear about regional and decadal failures.

I think I was looking for a bit more but on reflection your field is assessing performance not model development so it was unfair to expect you to stick your neck out in an area that is not your speciality.

I get the impression that the models tend to develop towards the minimum level of complexity necessary to do a reasonable hindcast and to keep the speed up so that finer resolution can be done. Is that your impression? Is model development on the right track or wrong track? OK it was not reasonable to expect you to put such things in a paper, but am I allowed to ask about your impressions about this sort of thing here?


Hank isn't main effect of SLR a direct impact on cities and society rather than a feedback where the effect needs to be modelled by a GCM? Don't you want a glacier/ice cap/ice sheet model(s) rather than a GCM? (Well perhaps a GCM to provide some of the inputs like water temperatures to the ice model.)

Steve Bloom said...

"Well perhaps a GCM to provide some of the inputs like water temperatures to the ice model."

That seems to be a notable area of failure at the moment, on both ends.

crandles said...

>"notable failure, on both ends"

Yes the GCM predictions are not reliable so what should we do.

Joughin, Smith and Medley (doi: 10.1126/science.1249055) seems to do a

If Thwaites Glacier continues to melt at the rate observed in 1996-2013 then according to simulations, it will collapse 200 to 300 years from now.

(Is that what was done or were some estimates of increases in rates used?)

That is likely to result in far too long a period before 'collapse' starts.

It seems to me that if you do that then it would be better to augment it with some other assumptions at least one of which will be wrong in the other direction. Perhaps more like:

Assume the melting rate increases with a delta temperature which is assumed to be smooth and a) remains constant, b) doubles every 50 years c) doubles every 20 years.

James Annan said...

Nick, it also talks about spatial scales. People certainly do present fairly detailed regional projections. But also, there have been plenty of claims of skillful short term predictions, not so much evidence though!

James Annan said...

Chris, I think that these days, people are busy putting in the latest fashion (eg carbon cycle, atmos chemistry) and can't afford to do too much with the physics and spatial resolution. But the topic was supposed to be confidence in climate models as they are, not a roadmap for how we should improve them.

The Old Man is back said...

Hi James,

A couple of thoughts come to mind. First is that the Global vs Regional issue is significant. A considerable amount of research and work goes in to regional projections, not least because these are of specific interest to large organisations and governments and so can attract funding or sponsorship. If, in spite of the progress on GCM's to date, there is still limited confidence in RCM outputs (and I think I understand why this is the case), aren't such efforts kinda pointless?

Which leads to the second point, which is that increasingly, there is a tendency to address climate issues on personal/local scales, because (probably rightly) many commenters and scientists perceive that a lot of the readers pay attention only as and when a specific personal impact comes into play.

Perhaps the time is coming where we start reminding people that this is after all a global issue, that worrying about what's going to happen to 'me' is really, truly less important than what's going to happen to lots of people, in lots of places.

Final question: in technical terms, how 'confident' does a result or a performance need to be for it to count as significant or meaningful? 75%? More? Less?

Steve Bloom said...

Fergus, IMO it's very helpful that we have a baseline of paleo data and non-model projections for context. Where I live in the US southwest, e.g.. it's pretty clear that without that context people would be much less concerned about drought projections from models. Of course they're not nearly as concerned as they ought to be, but that's a different story.

"Perhaps the time is coming where we start reminding people that this is after all a global issue"

Say what? I kind of think we've been doing that. A possible piece of evidence that there's been some (relative) success along those lines is the new U.S. survey data about differing responses to "global warming" and "climate change." My conclusion from it is that the long-term association between the (formerly more extensive) use of global warming and various (largely, at the time, *global* as stated) negative impacts (which, perhaps significantly, people were basically hearing about for the first time) actually sank into the public consciousness somewhat.

Steve Bloom said...

James, isn't increasing resolution largely dependent on the availability of new machines?

guthrie said...

There's nothing wrong with putting down in black and white stuff which 'everyone knows'; I've recently found that with some medieval archaeometallurgy stuff that everyone just takes for granted but nobody has actually proven one way or another. Until something is stated clearly it's hard to have a proper sensible discussion about it.

The Old Man is back said...

Okay, that was a pretty pompous- sounding sentence. Apologies for the clumsiness.
Completely agree about Palaeo - it's how I found James in the first place, after all :)

James Annan said...


Yes, but there is also a competition for the resource, and increased resolution isn't the sexiest idea around.

Fergus, yes well we started with global, but there is a big demand for regional predictions, and we can't really deliver them (yet).