Bishop Hill Lewis 2013 as an "outlier"
Apr 19, 2013 Climate: sensitivity One of the strangest things about Dana Nuccitelli's article about the Lewis 2013 paper is its twin-pronged attack - alleging that Lewis misrepresented the match between the mode of his estimate and that of Aldrin et al, and suggesting that the Lewis result is an outlier.
In the graph below, I have redrawn Lewis 2013 and a selection of other papers: Aldrin et al, Forster and Gregory, and Troy Masters' new paper. The IPCC's 2-4.5deg range is shown as the shaded area.
With Lewis's PDF sitting in the middle of the bunch, Nuccitelli's position is shown to be baseless. I've added a line through the mode of Lewis, which you can see matches Aldrin et al almost exactly. So when Nuccitelli says Lewis is misrepresenting the match between his findings and Aldrin's, it's easy to see that this is, ahem, a stretch.
;)
There are other papers that could be shown on the plot - Ring et al, Lindzen and Choi and so on, but I think this handful is good enough to make the point: there is simply no way Lewis 2013 can be seen as an outlier.
Reader Comments (61)
Uplier!
Looks like Nuccitelli is once again the outlier, possibly without the out.
It's SkS - what do you expect?
I second Martin A's comment.
High time to consider SkS as informative and authoritative as Roman Catholic's commentaries of Martin Luther, or a "History of the USA" written in North Korea.
Well, you know, they have to say something.
Nutty Jelly? No thanks, I'm Presbyterian.
omnologos pls.
Catholics were true to,their faith and totally scientific in comparison to Mann et al and SkS.
Am sure it is and will prove to be the truth.
Forever and ever...
Dominus vobiscum!
To be fair, Nic's probability density is quite different from the others in one respect. They all show some modest probability that the IPCC's best estimate (3.0) is correct; Nic disagrees. It would be interesting to hear him explain how this difference arises (presumably from the choice of prior).
Question: is Lewis an outlier? Answer: could he be? (His graph does go up a long way compared to the others, does it not?). So, what distinguishes the outlier? (I only ask because I don't know),
Lewis's pdf has been on a diet and doesn't have a fat tail. That makes it an outlier in today's world. It also explains the emphasis the team wants to put on the meaningless mean.
Possibly they could say the paper is an outlier in the degree of pointiness of the hump (technical?) But I noticed that Dana Nuccitelli didn’t adopt this graphical approach to show this, he seems to rather take the approach of a mangled word construction that can only be described by borrowing the words of Gavin Schmidt, as “sleight of hand” ;)
Of course once you see it graphically like this you can’t hide the fact that Lewis has a pointier shape and thinner tails, one can see the implications a lot clearer. The point is that Lewis's study is narrowing on a specific point that has precedence in other studies, only his does it with more certainty*. Doesn’t make it better or more truthful but makes it clearer.
*I admit possibly terminology faux pas here from my layman pov but I am only a visual biased kinda person.
Has anyone compiled a list/table of the mode and mean of the various PDFs out there? And what exactly is the significance of the mean? (Apologies for noddy question!)
Nic Lewis-
I would be most grateful if you could comment on the following. There have been persistent suspicions that apart from warming due to non CO2 related natural variability, the surface temperature instrumental records may have significant warming artifacts and bias due to under-corrected urbanisation UHI effects, station moves, station selection and station correction adjustments. If the surface temperature record ultimately turns out to have such bias, can we infer that the CO2 sensitivity would be still smaller?
What makes Lewis an outlier?
Further on, Nuccitelli writes ...
Is this an apples to apples discussion or apples to oranges?
Just to clarify, what is shown here is of course the preferred "main results" climate sensitivity PDF from my paper, which – following Forest 2006, on which it is based – does not allow for uncertainties in forcings other than aerosol forcing, or in observational surface temperatures. Aldrin 2012 does allow for these uncertainties, so it is unsurprising that its PDF is wider, with a lower peak. And IMO the use in Aldrin 2012 of subjective Bayesian priors, with a uniform prior for sensitivity, very likely leads (wrongly) to a wider sensitivity PDF than my objective Bayesian method would produce were it applied to the Aldrin 2012 data.
I state in my paper (without showing a PDF) that making allowance for uncertainties in forcings and observational surface temperatures widens the 5-95% uncertainty range for climate sensitivity from the basic 1.2–2.2°C to 1.0–3.0°C, with the median unchanged at 1.6°C. That corresponds to a PDF much closer to the Aldrin 2012 PDF in height and width - indeed wider than some of the alternative PDFs given in Aldrin 2012.
Has SS declared "The One Tree" as being an outlier too?
Pharos
"If the surface temperature record ultimately turns out to have such [UHI, homogenisation adjustments, etc, warming] bias, can we infer that the CO2 sensitivity would be still smaller?"
Yes, in principle. But bear in mind that these possible biases mainly apply to land measuring stations, and the ocean makes up 70% of the global surface. So even if the biases turn out to be substantial, their effect on the global mean temperature trend is unlikely to be huge.
FYI, over 1979-2011, when satellite data for temperatures in the lower troposphere are available, they show a trend increase of 0.14°C/decade in global mean temperature. That compares with 0.16°C/decade for the main surface temperature records (land: near-surface air; ocean: just below surface). The satellite temperatures won't be affected by the biases you mention, although they could have other errors and they measure something a bit different.
Nic
Most grateful for that.
Speed
"Is this an apples to apples discussion or apples to oranges?"
It is Nuccitelli , not me, who is misrepresenting the position and being misleading. One of his own regular contributors, Tom Curtis, takes him to task about this.
The Lewis curve is very symmetrical and not skewed. Poor use of priors was a main reason to generate artifical skewdness. Are there further reasons to give the symmetrical curve more plausibility ?
It would be interesting to evaluate the coefficient of variation ( = standard deviation / arithmetic mean) for each of the PDF's as a measure of how spread out it is about its mean.
This is an overall summary of a century's estimates of climate sensitivity to 2006 and some analysis, without confidence limits. It includes a comprehensive reference list.
http://bartonpaullevenson.com/ClimateSensitivity.html
There are about 50 papers by now, of which Bishop Hill has chosen four of the lowest for his graph. In the context of the full list, Lewis is near the bottom of the range.
It's not the most scientifically rigorous of lists, but the Wikpedia entry gives some feel for confidence limits.
http://en.wikipedia.org/wiki/Climate_sensitivity
BH was making the point that Nic Lewis's result was not an outlier. You don't have to be spot in the centre to qualify as "not an outlier".
The table of results you refer to is interesting.
A plot of CS against year does not show any apparent trend (eg for recent estimates to be smaller or larger than earlier ones).
I calculated the mean of the CS estimates in the table you refer to as 2.9, with a standard devation of 1.5.
This would also confirm a value of 1.6 as "not an outlier", being in the range (mean ± std dev).
Entropic Man,
the focus is on the best studies based on observational data.
Further, as we now today, many examples in your list are biased high due to an overstimate of aerosol cooling, underestimate of black carbon warming, lack of recent data and poor use of statistical methods.
"High time to consider SkS as informative and authoritative as Roman Catholic's commentaries of Martin Luther, or a "History of the USA" written in North Korea." --omnologos
Not a bad simile, but I lean towards TonyM's view. The Church tried for 30 years to obtain a formal reconciliation with Lutheranism. (A bit more recently, my Jesuit teachers told me Luther's 95 theses were mostly right.) A few years ago (few in the ecclesiastical sense), a new song appeared in the RC hymnal, "A Mighty Fortress is Our God," by some bloke named "M. Luther." This is like Mann mentioning Steve McIntyre in a footnote.
"Is this an apples to apples discussion or apples to oranges?" --Speed
Nuccitelli's comparison is applesauce.
Also interesting, that the probability of the previous central estimate (or higher) appears to be <1% or in IPCC terms "exceptionally unlikely.
One of the most helpful Bishop Hill threads of all time. Thank you, Andrew and Nic.
'Scooter-boy' has become the outlier ... his mates around him are starting to move towards a different climate orbit and he hasn't noticed the change going on around him.
Imagine a model where the only heating in the atmosphere is 23 W/m^2, and that solely via clouds where it is frequency shifted to go to space avoiding the CO2 dip, and reality is very different to the present Houghton/Sagan farrago......
It's probably worth adding Annan and Hargreaves 2012 to the list. They estimated ECS mode of 1.7deg C in range of 1.2 to 2.4 based on a re-evaluation of LGM temperature movement.
Nir Shaviv 2005 looked at a range of different geological timescales and found a range of 1-1.3-1.6 after including Cosmic Ray Flux as a regression variable. (1.6 to 2.4 if CRF is excluded.)
In fact, as far as I can tell, there are no well-founded observationally derived estimates of ECS which find a most likely value higher than 2.2deg C. (Please let me know if you find one.)
I think Nic is well inside the ballpark.
Thanks to Speed for providing the context: Nucitelli's outlier view is based on comparison with all estimates available,not just the four graphed at the top.,four at the low end of the spread.
@Apr 21, 2013 at 5:09 AM | Nick
Maybe I am missing some dry wit here, but I'm not sure of what you say you see there. Are you saying that the four graphed estimates above are all outliers together? I thought the concept outlier tended towards meaning rarer examples, if not singular. If there are "only" four there could you tell me what is the maximum number you would expect to see in a group of outliers - in this context?
Or is it some other persuasive aspect in the words of Nuccitelli, quoted by Speed, that help make the context clearer for you?
Nick -
Dana's post, beginning with its title, attempts to paint Lewis 2013 as a *single* data point, far from all other results:
I would say the existence of other studies with comparable results contradicts what one might say is an egregious claim.
Further, Speed's post to which you refer repeats Dana's mode-vs-mean comparison. The mode of Aldrin et al. is, as one can see clearly in BH's graphic, remarkably close to that of Lewis.
Nick Lewis probability density function is more constrained and has a much higher mode probability.
Does this mean that in his model, the data, the different data sets and the estimated model parameters support each other particularly well ?
Manfred -
Nic Lewis discusses this above in his comment of Apr 19, 2013 at 9:38 PM.
Looking at some of the recent papers, there's a pattern.
The low estimate papers are based on 20th century temperature records, a period of relatively uniform change over a limited range of both CO2 and temperature. The papers tend to assume that this linear relationship will continue and that climate sensitivity is a constant.. They also pay little attention to lag and presume that the immediate temperature increase is also the total temperature increase.
The higher estimate papers use more paleo data which gives information over a much larger range of both CO2 and temperature. pay more attention to temperature lags and take account of the possibility that the sensitivity may increase with time as positive feedback effects such as permafrost CO2 release kick in.
Entropic, See Annan's recent LGM paleo estimate. It came out late last year, 2012. Pretty recent I would say.
Harold W,while you claim Nuccitelli is trying to paint Lewis as a single data point, far from all other results you then immediately quote N saying 'outlier..lower..than most other studies'...that is an important enough distinction to note.
And the distinction between Lewis and Aldrin is clear in the PDF graphing. Plenty more of Aldrin lies above 2C. So N's position is hardly 'baseless' as in 'totally without basis'. Let's stop massaging words and phrases,eh?
entropic man: "take account of the possibility that the sensitivity may increase with time as positive feedback effects such as permafrost CO2 release kick in."
These are not new feedbacks! If these are real feedbacks, they have always been in play, and would have been part of any climate sensitivity over the past century or more.
And it's interesting that you mention "permafrost CO2 release". Because one of the things that happens as permafrost melts is that the vegetation starts photosynthesizing and capturing CO2. (I presume you meant CH4 release, but still...)
Nick
More of Aldrin's PDF lies above 2C because he uses a uniform prior in CS. I think this is the point that Nic L is trying to make. The mode is largely unaffected by choice of prior. It therefore makes sense for Nic to discuss it when comparing his results to other papers. Nic's mode is identical to Aldrin's, so to that extent they are reaching the same answer. But Nic has fixed the biasing influence of the uniform prior - that's the point of the paper - so he gets a much better constrained PDF, with less density above 2C.
Apr 22, 2013 at 4:59 AM | Nick
'outlier..lower..than most other studies'...
I love to see lectures about "massaging words" from someone wearing out their '.' key peppering their statements with ellipsis in one person statements, and implying they denote assent from someone else who was quoting them in full ;)
Low end of a "spread" of outliers, love it.
Yeah, yeah, I think we all know what Dana wants - he wants those highlighted words you chose - "lower", "outlier" "than most other studies" to stick out in our minds. But those words are meaningless on their own and rather have the misfortune to be attached to other words that in sum are "sleight of hand".
It is interesting now to see you say
My emphasis.
The graphing that you admit is clear denotes an explanatory visual feature, a visual feature overlooked by the "massaging words and phrases" from Dana in preference to give the impression of misrepresentation from Lewis.
If Dana had shown a comparative graph like the Bish does here then maybe his crap word usage could be forgiven as ineptitude, but as it stands it remains, as in Gavins words, "sleight of hand" ;)
Nick -
I am rather bemused that you accuse me of "massaging phrases". Just what part of Nuccitelli's "single-study syndrome" did you think that I massaged or otherwise mangled?
As to whether Lewis 2013 differs from the AR4 range -- well, I would say it is "not inconsistent with" AR4. If we reject all new values which show deviations from "consensus" values, simply because they differ, how is the state of knowledge to imporove?
I'm sure you're familiar with the history of Millikan's measurement of the electronic charge. I think it likely that ECS range of estimates will similarly gravitate towards lower values, as we incorporate more observational evidence. [Most higher estimates of ECS are based on GCMs; observational estimates have a history of being lower.]
And it's interesting that you mention "permafrost CO2 release". Because one of the things that happens as permafrost melts is that the vegetation starts photosynthesizing and capturing CO2. (I presume you meant CH4 release, but still...)
Apr 22, 2013 at 5:49 AM | Curt
That turns out not to be the case. The rate of CO2 release by decomposition as the tundra thaws exceeds the rate of CO2 uptake by growing vegetation. Warming tundra is a net positive feedback increasing CO2 as temperature increases.
http://www.terradaily.com/reports/Expansion_of_forests_in_the_European_Arctic_could_result_in_the_release_of_carbon_dioxide_999.html
CH4 is a separate problem, another positive feedback.
Curt
You assume that the feedbacks have had linear effect throughout. This may not be so. The Arctic minimum ice extent remained stable until the late 1970s and then went into a continuing decline. Past warming in Sibera produced a step change in the permafrost and may do so again, at around anomaly +1.3C.
http://www.ox.ac.uk/media/news_releases_for_journalists/130221.html
In a climate system which is chaotic rather than deterministic, non-linear behaviour is a more likely bet than steady change.
Most higher estimates of ECS are based on GCMs; observational estimates have a history of being lower.]
Apr 22, 2013 at 2:14 PM | HaroldW
With good reason. Our direct observational data only spans 120 years and a limited subset of temperature and CO2 content combinations.
Entropic man: "Our direct observational data only spans 120 years and a limited subset of temperature and CO2 content combinations."
GCMs have not been shown to have accuracy in projection even over a time period of a decade or two. [For shorter intervals, the uncertainty is such that their accuracy can neither be confirmed nor refuted.] Why do you think that GCMs are to be relied upon more than observation-based estimates?
HaroldW
You dont need GCMs. There's ice core data giving temperatures from oxygen isotope ratios and direct measurements of CO2 from trapped air. The glacial/interglacial cycle shows 9C temperatures and 200pm CO2 during glacial periods alternating with 14C and 280ppm CO2 in interglacials.
The orbital changes account for 1C and the rest is amplification as positive feedback, primary and secondary forcing between increasing temperature and increasing CO2 raise the temperature over the following 10,000 years. Thats an ECS of 4C for an extra 80 ppm or 10C for a doubling from 200ppm with no need for models or assumptions about feedbacks at all.
It's a natural cycle, not an anthropogenic one, so aerosols, urban heating etc are not there to complicate things. Its also way bigger than most climate scientists are willing to talk about in public.
Perhaps we are being too optimistic?
Entropic Man,
I am not sure if you are being intentionally funny or not. If you are, then forgive my lack of a sense of humour.
An increase in temperature causes an increase in atmospheric CO2 through outgassing from oceans. Let's say the rate is around 20 ppm per degC change in average surface temperature. A 4 deg C change in average surface temperature therefore gives rise to an 80ppm increase in CO2 and corresponds to a swing of about 9 degC in the subpolar regions where icecore data are typically collected.
Take one cart and one horse. Place in correct order. Then you can work out why most climate scientists aren't willing to repeat your calculation in public.
Entropic man -
10 K/doubling? I don't think I can talk you off of that ledge, if you think that's consistent with the ~0.5K increase we've seen from about half a doubling. But there are reasons why that's "way bigger than most climate scientists are willing to talk about in public"; a primary one is that attributing all forcing (except for Milanković) is a bridge too far.
There is one tipping point with high sensitivity between an ice age and an interglacial. If temperatures remain sufficently low during summer and snow doesn't melt glaciation on the northern land masses may start with strong positive albedo feedback.
Above current temperatures, there may be no similar point. If Arctic ice melts completely during summer, oceans take up more heat during day, but they also increaingly lose heat at night and from fall until refreeze, particularly heat transported to the Arctic from the south. Increased evaporation may also create more snow further south in fall, another negative feedback.
Antarctica is to cold to green.
That's why we had only 2 stable climates with continents around its current positions - ice ages and interglacials.