Bishop Hill AR5 inquiry followup
Mar 27, 2014 Climate: Allen Climate: sensitivity This is a guest post by Nic Lewis, describing the flurry of activity since he appeared before the Energy and Clmate Change Committee.
My comments on Myles Allen's oral evidence to the ECCC, and his response have been published.
Some things in Myles' response that might be worth pointing out:
1. Under Point 1: "The IPCC Summary for Policymakers does not give “best estimates” of 2100 temperature, largely because they would not be policy relevant: the one thing that can be said with confidence about best estimate predictions is that the real world will not follow them. A best estimate of a strongly skewed distribution is particularly misleading".
and under Points 4 & 5: "The IPCC does not give a best estimate of 2100 climate: it gives a range."
It may not call them "best estimates", but Table SPM.2 of the IPCC Summary for Policymakers (SPM) gives "mean projected changes" in global mean surface temperature (GMST) for 2081-2100 on each RCP scenario. And Figure SPM.7 graphs the CMIP5 multi-model [mean] simulated GMST to 2100 on the lowest and highest scenarios, as lines with a measure of uncertainty shaded. The means are implied to be best estimates by virtue of being highlighted as lines within the shaded range and by being the only central measures given. Moreover, Myles statement 'A best estimate of a strongly skewed distribution is particularly misleading' is very strange, since for every scenario the uncertainty distribution for GMST given in the SPM is symmetrical. The IPCC AR5 report does of course not give a point estimate for GMST in 2100 across all scenarios, which would indeed be misleading. Nor have I ever done so.
2. "Mr. Lewis argues that the IPCC’s ranges should be compared to his 17-83% ranges, which would suggest a larger discrepancy, but this would mean placing greater faith in his own analysis than any other study assessed by AR5. … If current evidence gives a 5-95% range for a forecast quantity, the IPCC typically assigns a lower than 90% confidence to the real world response falling inside that range".
I didn't actually make any such claim in my comments on his oral evidence or in any other evidence I provided to the ECCC. And TCR is not a 'forecast quantity' in this context. It is a property either of climate models, measurable with little uncertainty, or an estimate derived by scaling the historical temperature response of the climate system to increased forcing, based on uncertain estimates of both quantities.
3. "It is a common mistake to interpret these simple models too literally, such as arguing there is a “correct” heat capacity of the box representing the near-surface ocean."
What I objected to was Myles using the IPCC's 2-box model time constants, which given short and long term heat capacity elements of any fixed values are a function of ECS and TCR, with quite different ECS and TCR values from those used by the IPCC. Sure, the model is simple, but one should ensure consistency in using even a simple model. And while I agree there is no "correct" heat capacity of the box representing the near-surface ocean, if the combined heat capacity of the two boxes falls far short of the total ocean heat capacity – as it does in Myles' 2-box model – then the model cannot be considered physically sensible.
4. "Mr. Lewis argues that CMIP5 models are already converging towards equilibrium by 2100, making their 2100 temperatures more dependent on ECS than TCR."
I argued no such thing. I simply pointed out that the rises in CMIP5 model temperatures from now to 2091-2100 were actually more highly correlated with the model ECS values than TCR values, contrary to what one would have expected.
5. "The actual impact of prior choice on ECS and TCR inferred from the energy budget for the 2000s is straightforward to calculate from the numbers provided in Otto et al (2013), and clearly supports the view that prior choice is more of an issue for ECS than TCR."
I pointed out that prior choice was more of an issue for ECS than for TCR in my comments on Myles evidence; my objection was that he had said that the issue only applied to ECS. What I wrote was that the statistical controversy:
...particularly using uniform priors – applies to estimating TCR as well as ECS, albeit the effects are somewhat lesser for TCR. … Accordingly, Professor Allen's claim that the controversy about statistical methods revolving around ECS does not apply to TCR is wrong.
The example in my Appendix was as stated simplified and approximate, in that it only reflected the largest source of uncertainty. But it was not at all – as Myles claims – "entirely meaningless".
6. "The model Mr. Lewis is using assumes that future warming is given by future forcing multiplied by the TCR plus a small (0.15°C) adjustment to past forcing that is independent of both ECS and TCR. This is both physically incoherent (the adjustment to past forcing must also depend on ECS and TCR) and inconsistent with the results of more complex models."
The claim that I use an adjustment for past forcing that is independent of ECS and TCR is untrue. My warming projections were based on fixed best estimates for ECS and TCR, so there is no reason why the adjustment for past forcing (i.e., for emerging warming-in-the-pipeline) shouldn't be fixed.
Reader Comments (14)
Please keep comments polite and strictly on topic.
[Snip - O/T]
Dear Nic,
On best estimates, can I remind you that on 19th August last year, responding to the Final Draft of AR5, you said "I think that the decision not to give a most likely value for ECS (and indeed TCR) in AR5 was certainly sensible, given that GCM and observationally-based estimates now point to rather different values." Your main complaint now about IPCC seems to be that that it did not draw attention to best-estimates based on observational studies (since you agree with IPCC on the 5-95% range). It is these observational studies that give highly skewed distributions, and the interpretation of their best-estimates depends critically on the choice of statistical model. I know the interpretation of the best-estimate is the median of the posterior under your preferred approach, but yours is not the only one. Under a conventional frequentist interpretation, it is just the value that gives the best fit to these particular data, nothing more.
On your claim that your 17-83% ranges should be the ones to be compared with the IPCC's "likely" range, you make this claim explicitly in the GWPF report. TCR is the forecast of the response to a hypothetical scenario. It is not a fundamental physical quantity like the speed of light.
On the effective heat capacity of the ocean, here are a couple of lectures on this (sent to Andrew Montford -- I don't know how to post them), explaining (in the context of very simple models) why we would not expect the global ocean to warm up uniformly even in equilibrium, and hence why the combined effective heat capacity of a two-box model of ocean behaviour may well be different from the global ocean heat capacity.
On CMIP5, if it is true that the response to RCP8.5 is more highly correlated with ECS than TCR in 2100, then this would suggest that the slow adjustment time of many of the CMIP5 models is short relative to 90 years. I doubt this is the case — it certainly would not be indicated by the stabilisation runs. What I think your result probably reflects is the fact that TCR estimates in CMIP5 are noisier than ECS estimates — next time it would probably be a good idea to require initial-condition ensembles to estimate TCRs.
On your response to point 5, the shape of the Jeffreys prior on the distribution of a ratio of two normally distributed quantities depends on the ratio of their standard errors. In TCR, this ratio is necessarily larger than it is in ECS, which is why results with a Jeffreys prior are closer to results with a uniform prior in TCR than ECS. Obviously if you set this ratio to zero, as you did in your example, then you get a large, and meaningless, discrepancy.
On whether your 0.15 "correction" is independent of ECS and TCR (and, by the way, I still find it outrageous that you add a number that is less than internal variability to Jonathan and Piers' method and then claim yours is so original you not only don't need to acknowledge the source but go out of your way to rubbish their approach as "unscientific"), you told me you made the same correction for the lower and upper bounds of your projections and for the calculation of "effective TCR" from CMIP5 models. So you are considering this quantity as independent of ECS and TCR, which makes no sense. At risk of introducing you to yet another back-of-the-envelope climate model, the constant composition commitment by a given date depends on current forcing multiplied by the difference between the ECS and the TCR and a factor that depends on the slow adjustment time.
Regards,
Myles
Dear Professor Allen the problem I have with the "debate", as it is currently framed, is the "Official" assumption that CO2 is the primary driver of climate change. There is much observational evidence to suggest that models based on this assumption are flawed- e.g.
Fyfe, Gillett & Zwiers (2013) Overestimated global warming over the past 20 years. Nature Climate Change 3,767–769.
Humluma,Stordahlc, Solheimd (2013) The phase relation between atmospheric carbon dioxide and global temperature.
Global and Planetary Change; Volume 100, Pages 51–69.
Lindzen RS and Choi Y-S (2009) On the determination of climate feedbacks from ERBE data.
GEOPHYSICAL RESEARCH LETTERS, VOL. 36, L16705, doi:10.1029/2009GL039628
Rothman, D.H., Atmospheric carbon dioxide levels for the last 500 million years.
Proceedings of the National Academy of Sciences 99 (7): 4167-4171, (2002).
Moreover there is increasing evidence of other "drivers", so-called "natural variation". e.g;
Lu Q-B (2013) Cosmic-Ray-Driven Reaction and Greenhouse Effect of Halogenated Molecules: Culprits for Atmospheric Ozone Depletion and Global Climate Change. International Journal of Modern Physics B Vol. 27 (2013) 1350073.
K. Georgieva, C. Bianchi and B. Kirov(2005) Once again about global warming
and solar activity. Mem. S.A.It. Vol. 76, 969
S. Bal S. Schimanke, T. Spangehl andU. Cubasch (2011) On the robustness of the solar cycle signal in the Pacific region. Geophysical Research Letters Volume 38, Issue 14,
Scafetta N and West BJ (2007)
Phenomenological reconstructions of the solar signature in the Northern Hemisphere surface temperature records since 1600.J. of Geophysical Res., Vol. 112, D24S03, doi:10.1029/2007JD008437.
H. Svensmark, Martin B. Enghoff, Jens Olaf Pepke Pedersen (2013), Response of cloud condensation nuclei (>50 nm) to changes in ion-nucleation” Physics Letters A 377 2343–2347.
Davies, Molloy (2012) Global cloud height fluctuations measured by MISR* on Terra from 2000 to 2010. Geophysical Research Letters, Volume 39, Issue 3.
We are now been fed the "official" line that the 17 year "pause" is due to "natural variation"- as if the same "natural variation" could not have contributed to the pre-pause rise!
There are also increasingly ad-hoc assumptions been made that the "missing heat" is "hiding" in the deep oceans - which we "conveniently" cannot measure, but that this heat will come back to bite us-driving up global temperatures whilst simultaneously violating the first Law of Thermodynamics.
QB Lu, Scafetta and Svensmark are red queen territory. You can't believe them all at the same time. Very different mechanisms. Choose one and try and defend it. We can always use the giggles.
Don, Eli
This is O/T. We are discussing the Allen/Lewis exchange.
Eli, if you read what I wrote, I did not champion any mechanism, rather I made the point that there were alternatives, published in the peer-reviewed literature, to the CO2-does-it-all mechanism.
I also pointed out that the GC models, based on CO2 as the primary driver, have done a poor job of predicting observational changes.
The onus is on you and the IPCC to prove these alterrnatives wrong, rather than continue to "prove" CO2 as the primary driver. In this latter case you need to read Karl Popper.
Dear Myles
Yes, I don't think it would have been appropriate for AR5 to give a single central estimate for ECS (or TCR), since that is only appropriate for a unimodal distribution whilst in AR5 different principal lines of evidence point to different values. But, unlike you, I don't regard wide 5-95% ranges as by themselves giving sufficient information about a distribution.
It's not just me who thinks giving a median is appropriate - the AR5 ECS uncertainty bars give the median as well as the 5-95% range. If you don't accept that medians have a probabilistic interpretation, then you could give a 40-60% range instead. With the likely range given as well as the 5-95% range and the median (or a 40-60% range), a user will have a good picture of where probability is concentrated within the 5-95% range. And for ECS and TCR at least all these ranges will normally be similar whether frequentist of objective Bayesian methods are used – but generally quite different if a subjective Bayesian method is used.
You say it is the observational studies that give highly skewed distributions. And they should be so. The question is, why are the CMIP3 and CMIP5 ECS distributions nearly symmetrical? The answer is presumably that they don't really provide a valid probabilistic range in terms of the real climate system's properties. There was a good post at WUWT on this point: http://wattsupwiththat.com/2013/06/18/the-ensemble-of-models-is-completely-meaningless-statistically/.
Although TCR may be formally defined as the forecast response to a hypothetical scenario, if one takes the 'generic TCR' concept in AR5 then it can be estimated as the actual response to historical forcing. So I don't think a TCR estimate involves the same additional uncertainties that a forecast of, say, 21st century warming involves.
I agree that the effective heat capacity of the ocean may well differ from its nominal value. But whatever the effective total and quasi-mixed layer ocean heat capacities are taken as, the time constants of a 2-box model will vary with ECS and TCR, which is the point I make here.
You may well be right about CMIP5 TCR estimates being noisier than ECS ones. I, like you, don't believe that the slow adjustment time of CMIP5 models is short relative to 90 years.
The extent to which use of a uniform prior for TCR matters depends, roughly speaking, on the ratio of fractional uncertainty in temperature and forcing changes. Once that ratio gets much larger or smaller than one, further changes won't make that much difference. The full AR5 fractional forcing uncertainty is much larger than the fractional temperature change uncertainty, so whilst taking the latter as zero makes my illustration very approximate I don't think it makes it meaningless.
Lower and upper bounds for projections weren't given either in my evidence to the ECCC or in the Lewis/Crok report. Regarding the calculation of "effective TCR" for CMIP5, the amount of committed warming that emerges between now and 2090 is unaffected by whether a low or high sensitivity model is used to estimate future warming, and so is logically independent of the model ECS and TCR values. Given an estimated amount of 0.15 K, then for a model that estimates it at 0.4 K I treat the 0.25 K overestimate as forming part of the total future warming going into the effective TCR calculation. That is spelled out in the Lewis/Crok report.
You say 'I still find it outrageous that you add a number that is less than internal variability to Jonathan and Piers' method and then claim yours is so original you not only don't need to acknowledge the source but go out of your way to rubbish their approach as "unscientific"'.
As I explained before, I don't see my method as stemming from that in Gregory and Forster (2008), a paper that – being purely AOGCM orientated – I hadn't paid much attention to. Rather, I see the method as an application in reverse of the generic TCR definition given in AR5, combined with an adjustment to reflect emerging warming-in-the-pipeline. I don't really see that any acknowledgement is required for such a simple adaptation of the generic TCR concept. Neither do I claim originality for something so obvious. But I intended no disrespect to Jonathan Gregory or Piers Forster, and am happy to acknowledge their work in Gregory & Forster (2008). As you yourself first pointed out, their projection method is (unlike mine) not energy-conservative and therefore not strictly physical, but I'm happy to call it a simple empirical relationship rather than unscientific.
Regards,
Nic
Sorry, Don Keller is Gish galloping in the hope that something sticks. Pick one (the ones he has picked have been picked to death both on blogs and in the literature).
Dear Nic,
All the 1-sigma range (or 16-84% confidence interval) from a single study tells us is "we would be mildly surprised (at the one-in-three level) to obtain these observations if the true value of the parameter lies outside this range", which isn't very useful. Imagine the uproar if IPCC suddenly decided to switch to showing 1-sigma ranges from individual studies rather than 5-95% ranges. And yours is an individual study — just because you have used IPCC numbers, they still derive from a particular set of observations from a particular decade, which may not be representative of the true TCR.
I have had to piece together the basis of your claims from various sources, which is all the more reason you shouldn't make strong statements to policy-makers based on unpublished models. As you say in the footnote to which you draw attention, "These two definitions of effective TCR comply reasonably with the generic TCR definition in AR5 Section 10.8.1 provided of the order of 0.15◦C warming-in-the-pipeline emerges in the models." So you were assuming this warming-in-the pipeline was independent of the models' response. What you now say is that, by effective TCR, you actually meant some kind of mish-mash between the model's TCR and their adjustment to past forcing. But in that case, the fact that these "effective TCRs" are higher than the models' actual TCRs is not evidence of a further problem with the CMIP5 models, but just evidence that they conserve energy!
You claimed on Ed Hawkins' blog "I did not use for my projections the unscientific ‘kappa’ method used in Gregory and Forster (2008)." If you have a genuinely new and better way of making empirical projections, I strongly suggest that the right thing to do would be to write it down in a paper, test it out on a few cases where we know the answer (as Jonathan and Piers did) and submit it for publication, not to present it to the ECCC as superceding the CMIP5 ensemble. Based on the information you have provided, it seems that the only difference from the GF08 method is this 0.15◦C correction. While you are at it, please give uncertainty ranges — the fact that the GWPF report only gave "best-estimate" projections (and was silent on the standard 5-95% range on TCR) was very odd. Did the GWPF insist that you took them out?
Myles
Dear Myles
My effective TCR definition made clear that the deduction for warming-in-the-pipeline represented the estimated real-world amount, not each model's amount. That is appropriate for a measure relating projected future warming to the future forcing increase. But in any case I estimate the average CMIP5 model emerging warming-in-the-pipeline won't greatly exceed 0.15°C.
You ask re uncertainty ranges for projections: "Did the GWPF insist that you took them out?"
What an extraordinary suggestion! The GWPF didn't request any changes to the contents of the Lewis/Crok report, or have any input to them.
The projections draw attention to the large difference between where model-based and observationally-based methods are pointing for the rest of this century. A comparison of central estimates is most relevant for that purpose. A crude uncertainty range for the TCR-based projections can readily be derived from the 1.0–2.0°C TCR 'likely' range we give, by scaling them down by a factor of 1.00/1.35 and up by a factor of 2.00/1.35.
Nic
A comparison of central estimates with no assessment of uncertainty is utterly meaningless. I'm surprised this point needs to be made on Bishop Hill. Anyway, I think this thread has run into the sand, so over and out.
Nic and Myles,
It is a pleasure to read a discussion where actual numbers and serious ideas are discussed with a minimum of name-calling and invective. It has been most informative. Thanks!
Your discussion makes me hopeful that we may, at last, be starting to make some progress on converging to a better understanding of climate sensitivity. A little off topic, but a related issue I'd like to encourage you and others to become more interested in is the reality of the various alternate scenarios. Many of the "how bad can it get?" discussions refer to the "hot" rcp85 scenario. It is my observation that this scenario probably assumes the burning of far more fossil fuel than can reasonably be economically extracted, given the earth's finite fossil fuel resources. Others that have looked into it, not all "skeptics", agree. The subject is important because scenario differences can contribute as much or more to projected long-term warming differences as can climate-sensitivity differences. This is another issue that will not be easy to resolve, but it certainly demands that good people begin to consider it seriously.
Professor Myles this very recent paper:
Trajectory sensitivity of the transient climate response to cumulative carbon emissions
http://onlinelibrary.wiley.com/doi/10.1002/2013GL059141/abstract
DOI: 10.1002/2013GL059141 Published in Geophysical Research Letters.
Shows a much lower TCR than the IPCC, irrespective of whatever "range" you chose to employ.
"We find the range of TCRE resulting from varying emission pathways to be 0.76 to 1.04 °C/TtC"
P.S. Real "Red Queen" stuff, Eli!