Tuesday, November 08, 2011

Schmittner on sensitivity

Yet another interesting paper, this time on climate sensitivity estimated from the Last Glacial Maximum. What makes this particularly novel and significant is that they have used two recently-developed and rather comprehensive spatially-resolved data sets, for ocean and land temperatures respectively, rather than relying on large spatial averages that most people (including myself twice) have relied on in the past. They conclude that sensitivity is "likely" to lik lie in the range 1.7-2.6K, very much towards the low end of most estimates and with very low uncertainty.

A weakness of the paper, however, is that the authors may not have adequately considered nonlinearity in the equilibrium response of the climate system to different combinations of negative and positive forcings. A number of papers (eg here, here and here) have shown that the degree of nonlinearity can vary significantly between different models, and although I have not used the energy-balance style model that Schmittner et al use, I suspect it will not represent this range of uncertainty well. What this means is, that even though they may be able to accurately estimate the "sensitivity" at the LGM, in terms of the ratio of temperature response to net radiative forcing, we cannot be sure how this will translate into "sensitivity" for 2xCO2. A possibly more statistically sophisticated and comprehensive attempt to account for uncertainties can be found here, for example.

That said, it's a useful antidote to the exaggerated uncertainty estimates that have been prevalent over recent years, and I certainly applaud the intentions and effort underlying this substantial piece of work. In any case, I expect the merchants of doubt to do their worst on it when they cite it in the IPCC report.


Nick Barnes said...


crf said...

How long do different models give to come to an equilibrium? Is there a lot of difference between different models?

James Annan said...

I think it's roughly similar, a couple of hundred years gets you most of the way there. Until recently, people didn't like to do such experiments, preferring to use a "slab" ocean which as well as being computationally cheaper in itself, also converges in a few decades (but which does not allow for changes in ocean heat transport). The last glacial maximum is generally taken as the mean state over a 4000 year period, it wasn't a true equilibrium (there were changes in that time frame) but was reasonably stable in terms of the difference from the present day. Makes life interesting when you have pollen data available which apparently lies under the edge of the (maximum) ice sheet extent!

In theory, higher sensitivity models should take longer than low sensitivity models to converge, other things being equal, but this isn't a huge effect across the IPCC model spread.

Ned said...

I admit I haven't delved into this deeply, but if I'm understanding this correctly, the temperature reconstructions they're basing this analysis on show a relatively small (compared to previous estimates) change in temperatures between the LGM & Holocene.

Most people would agree that the environmental consequences of the change from LGM to Holocene were non-trivial. (My house would have been under a kilometer or two of ice, so it's a pretty significant change to me, but YMMV).

But does it really tell us anything new, if the old studies said that the temperature change from the LGM-Holocene transition was X, and climate sensitivity is Y, while this new study says that the temperature change was X/2 and climate sensitivity is Y/2? Projected temperatures in 2100 would be lower than if you used a sensitivity of 3C, but the planetary environment would be more sensitive to small temperature changes. So it would be more or less a wash.

One could achieve a similar effect by switching one's units from degrees F to degrees C. The numbers are smaller but the end result is unchanged.

Of course, I might be misunderstanding something here.

James Annan said...

That's an interesting way to spin it, I hadn't really thought of it in that way. It's true that (assuming linearity) 4xCO2 is of similar magnitude to the LGM change, irrespective of the actual temperature changes they both represent.

But I think most would say that the lower temp change actually means that the LGM was less radically different than we previously thought, not that a smaller temp change was more important than previously thought. Of course areas that got covered in ice did see a radical change, but that is only a small part of the globe. And impacts are generally calibrated to absolute temperature changes, not fractions-of-an-LGM. But if more radical environmental changes were linked (historically) to smaller temp changes, then this might be grounds for changing the calibration...

Davo said...

I'm not convinced the novel aspect of the data sets is that they are spatial more that they are radically different in magnitude from previous studies. I note from your second paper that you used a value of -6K and Shakun & Carlson (2010) give a value of "at least" -5K. These guys have come up with an estimate around -2.5K which is going to halve any sensitivity you come up with.

James Annan said...

Well our estimate at least was a bit of a wild guess rather than the result of a proper fit to data. But it is interesting to see how estimates have come full circle: the CLIMAP data set (~1980) showed little cooling, then around the time of the IPCC TAR and AR4 many people preferred alkenone proxy data that gave a bigger result, and now MARGO is close to CLIMAP again. (That's all ocean data, I haven't looked at the land stuff much.) It is certainly interesting to see how fashions change...

Hank Roberts said...

> fashions change

I'd think the ocean scientists would comment on how limited the samples taken from the fossil record are thus far, compared to the known variability in today's ocean. What would happen, say, if someone were to take a limited number of samples out of the contemporary knowledge about ocean conditions --- picking only a few, only to the same extent that fossil studies sample the conditions of the paleo ocean?

Sample the contemporary record in a way (distribution across the map?) (depths/temperatures/biota in the current ocean comparable to those found in the strata from which the known fossil record was laid down?)

If we took from the current total known ocean data set only samples that would get about the same kind of info as is used for each of the different paleo studies --

How different would our view of the contemporary ocean look, if all we knew about it was as much as is used for the big paleo studies?

poorly written question, hope it's clear enough. There must be a statistical procedure for this kind of thing, maybe already being done? Do the same probe several times on a known variable data/population and on the unknown, and figure out some error range?

Rattus Norvegicus said...

The peddlers of doubt aren't even waiting for the IPCC to cite it:


Chip Knappenberger said...


Maybe I have misinterpreted who James is referring to as the "merchants of doubt" in this case. I know who Oreskes was referring to, but, I got the feeling that James was taking aim at his doubters in the IPCC.



Ned said...

Thank you for the reply, James. But I'm somewhat dubious about this:

"the LGM was less radically different than we previously thought [...] Of course areas that got covered in ice did see a radical change, but that is only a small part of the globe. "

Well, coastal areas worldwide were certainly affected by the 120 m rise in sea level. Pretty impressive if that much SLR can be caused by a 2.2 C warming.

My admittedly nonexpert impression is that there is quite a bit of evidence for shifts in ecotones, changes in precipitation regimes, etc. that would certainly be considered significant by the average farmer in many non-ice-covered regions.

I wouldn't be particularly affected by a paper that said "oh, the climate sensitivity is lower than we thought, but the ecosystem sensitivity to small changes in global temperature is higher than we thought, so it all cancels out."

In contrast, I would be rather surprised to learn that the LGM Earth was basically just like today's Earth but with more ice.

Richard James said...

The link to the paper is broken - perhaps someone worried about embargos being breached. Does anybody know where another copy of the paper can be found?

crandles said...

Link works for me. You could try


James Annan said...

Thanks Chris - interestingly, Andreas told me last week it was on his website, but it isn't, so I guess someone got cold feet - this is going to be in Science soon.

Chip, you read me right. I must admit, I'm a little surprised at your lack of scepticism over the paper (as per the WCR post).

Hank, I think the argument is (in part) between different types of proxies, and their interpretation - not actually an observational uncertainty as such (though to the end user, it looks like that). How the proxies respond to different climatic conditions is always open to question...

James Annan said...

Chris, do you have a similar link to the supplementary info?

Chip Knappenberger said...


Well, the WCR coverage was more of an announcement of the findings rather than a critical review—which could be found if any of our readers followed the link to your article :^)

Am I mistaken that there seems a tendency for recent papers incorporating more paleo-data to produce less of a fat tail (and perhaps even a shift of the median value towards the left) than indicated in the IPCC AR4?

(I’ll admit, however, that I was surprised by how little colder the LGM was than the present in the Schmittner et al. paper—a key part of their low sensitivity result)


David B. Benson said...

It wasn't too cold for humans during LGM in even some rather surprising places although the best archaeological data seems to be from around 14 kya and later. But irrespective of the temperatures in various regions, globally it certainly appears to have been quite dry:
LGM vegetation map:

Richard James said...

Thanks for the link to the paper - looking forward to reading the supplementary material to find out how they deal with the uncertainty in the palaeo data. At least some of the MARGO data is rather dubious (large positive anomalies in the Norwegian Sea), but perhaps not enough to seriously bias the analysis.

James Annan said...

Found the SOM on the google cache. I remain particularly unconvinced about the highly multimodal nature of the result, for which I can think of no good physical justification. The authors suggest it may be partly an artefact.

Eg, simple arguments suggest that a sensitivity of (x+y)/2 should normally give a better fit to the data than the average of the fits of x and y, for any x and y - this is just the same thing as the mean of an ensemble (of 2!) being better than the ensemble members, about which I have written quite a lot recently...

Another thing that seems curious to me is their use of elevation (only) to indicate the larger ice sheet, relying on snowfall to generate the correct albedo. However it's not clear how this would affect the results.

PeteB said...

To be honest I always thought that the observational estimates gave a slightly lower estimate of CS (although still within the IPCC range)


Summary of Observational Constraints for Climate Sensitivity

....Results from studies of observed climate change and the consistency of estimates from different time periods indicate that ECS is very likely larger than 1.5°C with a most likely value between 2°C and 3°C.

crandles said...

Google Webcache worked for this


googledocs had a permission/unable to retrieve issue.

chriscolose said...

There is a lot of literature lately on the best way to estimate climate sensitivity (whether we use perturbed physics ensembles, or in the case of people like Hansen, simply determine the best forcing & T anomaly estimate and assume an invariant climate sensitivity parameter between states). The latter approach seems to have fallen out of favor a bit, but I'm not so sure the assumption is any more problematic than assuming that the change in relevant feedbacks between the two states are well simulated by the model (as in the perturbed physics ensembles).

I think it's an improved way of thinking about the uncertainty, but it still seems to me that some of the largest issues with the LGM are the relatively uncertain estimates of the total forcing, which don't seem to have been fully probed by most studies (Kohler et al 2010 provide one of the best summaries of all the relevant factors). This is particularly true on the shortwave (albedo) side of the energy budget, where data constraints are sparse for some of the factors we know should be important.

I think people are going to have a hard time arguing for an LGM-Holocene temperature difference of significantly different than ~5 +/- 1 K. Something on the scale of -2 K temperature difference is almost certainly incorrect. Unfortunately we don't have a long-term time series of the global temperature anomaly, forcing people to assuming simple scaling relations between Antarctica (or the deep ocean) and the global mean, which themselves are almost certainly climate dependent.

But I'm glad there are more studies recently investigating the differential feedback structure between states, which should allow us to better understand how climate sensitivity varies as a function of the base climate.

Magnus Westerstrand said...
This comment has been removed by the author.
Dallas said...

It appears that nonlinear climate sensitivity is as certain controversial leaks :)

When do you think better communication of the southern, tropical and northern predictions instead of not very useful global should start making the news?

It looks to me land use changes may be a bit underestimated. The energy models should be better for estimating that.