The long-awaited prequel to "Can the Last Glacial Maximum constrain climate sensitivity?" is now up on CPD and open for comment. This latest episode is set back in the mid-Pliocene warm period 3-3.3Ma BP (MPWP, sometimes also called mid-Piacenzian by people who care deeply about the conventions for paleoclimate nomenclature). The particular reason for looking at the MPWP is that it’s the most recent period when CO2 was relatively high (at least compared to pre-industrial, though perhaps not so high compared to today) and the climate was correspondingly warm. This means that it avoids some of the major complications that arise when studying the Last Glacial Maximum, where there is the question of how directly a strong cooling informs about a strong warming, and large ice sheets provide a substantial forcing to the atmosphere-ocean system that does not necessarily combine linearly with other forcings such as GHGs. Thus, PlioMIP was born around 2009, and by 2015 lots of different modelling groups had produced simulations of the MPWP to compare with each other and with the data assembled by the USGS. Being much further back in time than the LGM, the data are much more sparse (even though the interval of time is also far broader) and more uncertain. Also, the boundary conditions such as atmospheric CO2 level are not known with great precision. One might be suspicious that choice of value used in the simulations, 405ppm, could have been been influenced by political considerations :-)
Although theres been a fair bit of analysis of the model results, no-one had looked directly at how the simulations depended on the equilibrium climate sensitivities of the models. So we did the usual thing, of comparing the tropical temperature change at the MPWP, to the equilibrium sensitivity of the models. It looks like there might be some sort of relationship there, though it’s far from certain.
Here’s the main result, with the model results represented by red dots.
The temperature data for the MPWP is a bit less well developed than for the LGM, so we didn’t really have a good uncertainty estimate but instead just tried a few possible values as a sensitivity test. It is also notable that the model ensemble seems to generally show more warming here than the data, although with substantial overlap if uncertainty is taken into account. It is possible that the forcing is too strong, or that the models are too sensitive in this region, or it may be that the data are wrong, or to be more precise, misinterpreted in terms of climate. I try to be avoid saying things like "the data are wrong" especially when talking to the data specialists, as at some level the data are definitely correct, real stuff has been measured and the measurements themselves are not in any real doubt. It’s how these real measurements are translated into a putative climate variable that is where the difficulties lie. Another interesting and possibly disconcerting aspect of this result is that the regression line does not pass through (or close to) the origin, implying that a hypothetical model with zero sensitivity to CO2 would show a significant tropical cooling at the MPWP. It is hard to see how that might arise, although it is also hard to see how a model with zero sensitivity to CO2 could arise anyway. It is perhaps not completely impossible that increased latitudinal transport would result in a tropical cooling accompanied by a high latitude warming so as to cancel out the change on the global average. After all, there has to be some explanation as to why a small change in model sensitivity seems to correspond to such a large change in MPWP tropical warming in this sample. We've presented this work a few times and no-one has come up with any great ideas about it.
Anyway, we don’t have any strong conclusions, it might be possible to generate a meaningful constraint out of this, but we don’t have any real confidence in the specific result presented here. PlioMIP2 is being planned, with various improvements over the original PlioMIP. Most importantly perhaps, the researchers are aiming for a true "time slice" in which orbital and other forcings are near enough constant (and so, hopefully, the climate was in some sort of equilibrium with this), rather than aiming for some sort of average over the warm peaks within a longer interval, as PlioMIP did. The main problem with this is that the data become even more sparse, but it should in principle enable a more meaningful model-data comparison. Watch this space!
20 comments:
At last, the catnip I wanted...
Thanks!
I'm watching.
With regard to that figure there are at least three interesting issues.
1. There was an unknown forcing or set of forcings at the time which overwhelmed the ghg forcing. Probably not but you have to consider it.
2. The data sucks. While obviously not the best fit, a forced fit through zero does not give a much different answer for sensitivity and benefits from physicality if you reject 1
3. The models suck, so their estimates of sensitivity are of not much use.
Eli thinks you have to abandon that fit or at least better justify using it,
Looking at the paper, it seems that the range is something like 1.8 - 3.6K. My understanding is that GCMs that do a good job of matching certain emergent constraints (my terminology may be poor here) suggest an ECS > 3K. Do you have any thoughts on that?
Eli, yes to be honest we just did the regression witnout thinking too hard, when we did it for the LGM the line actually did go close to the origin. It was only later that we realised it was a bit odd. We are planning to put the no-constant version in the paper too.
ATTP, yes there's a hint of tension there with the paleo and recent obs tending to point to a slightly lower value compared to the emergent constraints, though it must be said that there's also plenty of overlap. It's not really clear what the uncertainties are in the emergent constraints work anyway. And we have reason to think the recent obs stuff is artificially biased a bit low. It's an interesting question as to how to combine all these bits of information...something I hope to make another contribution to at some point.
Thanks. Your point about there being a lot of overlap is well made. It would indeed be interesting to understand better how to combine all of this.
ATTP, you say "GCMs that do a good job of matching certain emergent constraints (my terminology may be poor here) suggest an ECS > 3K". I suspect that you are thinking particularly of the Sherwood et al 2014 study. However, the Zhao et al 2016 paper , from a sober GFDL team, casts severe doubt on its validity, finding the opposite relationship to that in Sherwood et al across several different versions of GFDL models with differing ECS values. Moreover, Zhao et al found that ECS could be engineered at will within a wide range, simply by varying the parameterization of convective precipitation, with no clear constraint that made one model choice more plausible than another.
James, you say "And we have reason to think the recent obs stuff is artificially biased a bit low." May I ask what is the evidence that you are thinking of here?
"sober"
Just can't resist working in the propaganda, Nic?
Here's another wrinkle for you.
But as the topic is the MPWP vs. current climate, it's odd you're neglecting the well-known key point that earth system models continue to be unable to manage a transition from the latter to the former. What can it mean?
Also: "simply by varying the parameterization of convective precipitation"
Er, simply? Yes, who'd have imagined that changes in a little thing like convective precipitation could have such an effect.
Steve Bloom,
Actually, "sensible and sober" was the description of the GFDL group's statements, in email correspondence, by one of the top scientists in this field.
"simply by varying the parameterization of convective precipitation"
You miss a key point, which is that no other parameters had to be retuned to maintain a realistic simulation of the current climate.
James,
I like the idea of generating a constraint on ECS from Piocene data, but as you know I have some serious reservations about some of your data and methodology, and results. I hope you won't mind if I follow up on some of these issues here.
1. In Julia and your response to my interactive comment on your Clim. Past Discuss. manuscript, in relation to my disputing the ECS values used for two models, you wrote in relation to IPSL-CM5A (where you used 3.4 C and I quoted statements from the model team that its ECS was 4.1 C or 4.0 C):
"the value of 3.4C quoted in Haywood et al appears to match closely to the 2xCO2 value of 3.47C presented in Table 1 of Dufresne et al (whether the small difference may be due to a rounding error, or some other source of variation is not clear). Thus, we see no basis for changing this value."
How can you possibly justify this? For a start, the 3.47 C value in Table 1 of Dufresne et al is for the IPSL-CM5A-MR version, whereas it was the IPSL-CM5A-LR version that participated in PlioMIP (check the resolution per Table 1 of Haywood et al). The value for IPSL-CM5A-LR is 3.59 C, not 3.47 C.
Secondly, and more significantly, 3.59 C is the effective sensitivity derived from the average climate feedback parameter during years 56 to 85 of the 1 %-per-year CO2 experiment. It is not at all usual to estimate an AOGCM's equilibrium climate sensitivity in this way, as I am sure you are aware.
Table 1 of Dufresne et al also gives the ECS value of 4.1 C from regressing over the 150 year abrupt 4xCO2 experiment, which is what I quoted. As I'm sure you know, that is the current standard method of estimating ECS in AOGCMs. The CMIP5 model ECS values given in IPCC AR5 all use that method.
Moreover, is clear that Dufresne et al regard 4.1 C, not 3.59 C, as the correct estimate of the ECS for their model. As I commented, they go on to say "… the climate sensitivity of IPSL-CM5A-LR ((ΔTse (2CO2) ~ 4.1K) lies in the upper part of the sensitivity range of the CMIP3 models,… ".
I don't see how you can justify scientifically the reuse of a model ECS value from Haywood et al. which is clearly well out of line with what both the model team and the IPCC authors consider to be the correct value.
2. In your response to my comment, Julia and you also wrote:
"even when changing both the GISS values to 2.3 and the IPSL value to 4.1 as suggested in your comment, the regression is still significant at the 5% level, contrary to your assertion."
On my calculations, based on careful digitisation of the tropical SAT anomaly points plotted in your Figure 2, the significance level is only 9.3%. Please would you state the nine tropical SAT anomaly values that you used, so that we can resolve this difference.
3. On my calculations, performing the regression without a constant term will reduce the median ECS estimate to 1.3 C whether or not either or both the disputed model ECS values are changed. Do you agree?
The comment was more about how you slipped it in. I'm sure your source is top indeed, or at the very least someone who agrees with you a lot.
No, I didn't miss that point, although I've only read the abstract. It's interesting, and we'll want to see what other sober and sensible scientists will make of it (if any can be found to meet that high standard). In and of itself it doesn't seem surprising that results would be pretty tightly clustered in the present compared to the future.
Speaking for my non-scientist self, while I remain a big fan of improving models to the point where they can inform more usefully about our trajectory into the future, I tend to take Hansen's advice about preferring current obs and paleo data to models. While I suppose the observed rapid changes to the climate, particularly in the Arctic, aren't a guarantee of a fast transition toward and beyond an MPWP-like climate state, it's clear enough that they're entirely consistent with one.
Nic,
I'll have a look at the Zhao et al. paper.
James, you say "And we have reason to think the recent obs stuff is artificially biased a bit low." May I ask what is the evidence that you are thinking of here?
I'll post a comment here that I've just made on my blog. From memory, the EBM best estimates for TCR are somewhere between 1.4 and 1.5C, while EBM best estimates for ECS are somewhere between 1.6 and 1.7C; a TCR-to-ECS ratio of 0.8 to 0.9. Currently (last decade or so) we have a planetary energy imbalance of around 0.6W/m^2 (maybe a bit higher). Given a Planck response of 3.2W/m^2/K this implies a minimum committed warming (given that even EBMs suggest the non-Planck feedback response is positive) of around 0.2C. Given this, how can the difference between the TCR and ECS be as low as 0.2C?
Also, given how we've warmed, and our current change in anthropogenic forcing, is there much chance that the TCR can be lower than 1.4C? I would think not, but maybe you can present an argument as to how this might be possible.
I'd also be interested in James's thoughts on the above.
So, yes, I think there is evidence that observationally-based estimates are biased low. I also think that it's worth acknowledging that they have inherent assumptions (linearity, for example) that may not be true (even if the apparent non-linearity is to do with spatial variability, rather than an actual non-linear response).
Take a look at Technical Note: Calculating state dependent equilibrium climate sensitivity from palaeodata
Peter Köhler1, Lennert. B. Stap2, Anna S. von der Heydt2, Bas de Boer3, and Roderik S. W. van de Wal2
http://www.clim-past-discuss.net/cp-2016-23/
Noodling around the EGU abstracts, I find the interesting "Sensitivity of Pliocene Arctic climate to orbital forcing, atmospheric CO2 and sea ice albedo parameterisation", produced a mere double marathon or so distant from you (Haywood again) and taking the opposite approach to the time-slice one. Presumably they can manage that because they're just focused on sea ice rather than the whole climate, and sea ice does seem like it should be central to resolving the Arctic amplification shortfall in the models. The upshot:
"Compared to a mPWP control simulation, monthly mean increases north of 60◦N of up to 4.2◦C (SST) and 9.8◦C (SAT) are simulated."
That refers to HadCM3, I guess. It's a big increase (is HadCM3 actually that bad in the Pliocene?), but they say that even so the problem has not been solved:
"However, data-model comparisons show the model temperatures still fail to match the proxy data temperatures."
Then they say, shades of that recent heretical cloud sensitivity paper:
"It is suggested that further high latitude warming may be achieved through adjustments to cloud parameterisation, although the gap between model and data temperature in simulations, even with significantly reduced sea ice cover compared to the control, suggests that agreement may still be difficult to achieve."
So still missing a heat transport mechanism?
Thoughts?
Too knackered from flu to give a coherent response but will try to have a look at that!
Prompt. ;)
I'm seriously curious about your view (singly or doubly) of this. Did you perhaps have the chance to talk to any of them in Vienna?
Shifting gears, oh look, another one. From the abstract:
"These low observational EfCS/two-zone EBM values have been questioned because (a) they disagree with higher observational EfCS/ZDM values, and (b) the EfCS/two-zone EBM values given by GCMs are poorly correlated with the standard GCM sensitivity estimates."
Hmm, nothing at all about the paleo mismatch. Imagine that.
Saw your to-the-point comment on Bates quoted over at ATTP's (possibly from here somewhere, and if so I probably saw it but forgot). I suppose we'll never know what the editor and peer reviewers were thinking. Perhaps the approach was novel, but if it's implemented wrong in the paper then why the green light? Andy Dessler points to extensive publication-shopping, so I suppose persistence has its rewards. ICYMI Bates was Jule Charney's grad student, which is poignant.
Post a Comment