Sans science, sans maths, sans everything
Jan 12, 2015
Bishop Hill in Climate: sensitivity

Over at Lucia's place, there has been some interesting conversation in the comments about the technical abilities of some of those on the other side of the climate debate. This originally arose in connection the host of And Then There's Physics, who had apparently reduced Blackboard regular Paul K to laughter in a post about a paper on climate sensitivity by Craig Loehle and a response to it, Cawley et al, which was written by five of the denizens of Skeptical Science. This amusement was followed by others chipping in with their own surprise at ATTP's comments. It's all good family fun. However, it turns out that it's not only ATTP who is struggling. Nic Lewis has added a comment to the thread about the Cawley paper itself which is astonishing.

First a bit of background. Loehle's paper described a model developed in an earlier paper (Loehle and Scafetta 2011) and used it to derive an anthropogenic warming trend from the mid-20th century onwards. The paper then derived an estimate of transient climate response (TCR) of 1.1°C by relating the anthropogenic trend rise in temperature to the increase in anthropogenic radiative forcing, and went on to derive an estimate of equilibrium climate sensitivity (ECS) of 2.0°C. The Skeptical Science response appeared about six months later and was written by Gavin Cawley, Kevin Cowtan, Robert Way, Peter Jacobs and Ari Jokimäki. This is the key part of the abstract.

We demonstrate that the Loehle and Scafetta model systematically underestimates the transient climate response, due to a number of unsupportable assumptions regarding the climate system. Once the flaws in Loehle and Scafetta’s model are addressed, the estimates of transient climate response and equilibrium climate sensitivity derived from the model are entirely consistent with those obtained from general circulation models, and indeed exclude the possibility of low climate sensitivity, directly contradicting the principal conclusion drawn by Loehle. Further, we present an even more parsimonious model for estimating climate sensitivity. Our model is based on observed changes in radiative forcings, and is therefore constrained by physics, unlike the Loehle model, which is little more than a curve-fitting exercise.

Nic notes that one of the key issues that Cawley et al raise concerns Loehle's estimate of the forcing that has produced the recent warming:

The paper notes...that Loehle assumed aerosol and non-CO2 greenhouse gases and other forcings approximately cancel each other out, and accordingly used only CO2 forcing. The Cawley et al. authors dispute the validity of this assumption, saying about the IPCC AR4 chart of 1750–2005 forcings that Loehle cited in support of it:

This does not however imply that these forcings have approximately cancelled over the period from 1951 to 2010, used to estimate the anthropogenic warming after 1950. The RCP8.5 forcings (Meinshausen et al., 2011), shown in Fig. 3(a) suggest that total anthropogenic forcing since 1950 has risen appreciably faster than the forcing from CO2 alone by a ratio of approximately 1.145:1.

Nic agrees with this this point, and in fact goes on to note that even Cawley et al's figure is too low.

Cawley et al. are correct in their assertion that total anthropogenic forcing since 1950 has exceeded that from CO2 alone. However, they understate the difference over the stated period, for two reasons. First, 1.145 is the ratio of the RCP8.5 increase between 1950 and 2010 in total forcing (including solar and volcanic), not anthropogenic forcing, to that in CO2 forcing. The correct ratio is 1.245.  Secondly, the RCP forcings dataset does not represent current best estimates. Based on the IPCC AR5 forcing dataset, the ratio of the increase in total anthropogenic forcing over 1950–2010 to that from CO2 alone is 1.361 times.

But this is where it gets rather hilarious. In simple terms, transient climate response is calculated as follows:

TCR = Temperature change/change in forcing

So if, as everyone agrees, Loehle got his forcing too small then his TCR figure is actually too large not, as the crack team at Skeptical Science seem to think, too small as well.

I was rather nonplussed at this point in my reading of the Cawley paper. Its main thesis was that Loehle had underestimated transient climate response (TCR) and hence equilibrium climate sensitivity (ECS). But Loehle derived TCR by dividing the estimated anthropogenic warming by the estimated change in anthropogenic forcing (relative to that from a doubling of CO2 concentration) over 1959–2013, which is broadly consistent with the generic definition of TCR given in AR5 (10.8.1). Obviously, if the change in forcing were greater than Loehle had assumed, that would imply an overestimate of TCR and thereby of ECS, not an underestimate. Very odd.  But then I read on:

As a result of this assumption, the method of LS14 underestimates climate sensitivity by about 13%...

Unbelievable! Instead of adjusting the Loehle TCR estimate by dividing it by 1.145, to reflect Loehle's underestimate of forcing by that ratio, Cawley et al. have multiplied the sensitivity estimate (which is for TCR here, not ECS) by 1.145. On that incorrect arithmetical basis, Loehle's method of working from just the increase in CO2 forcing would indeed have underestimated TCR by 13%. But the correct conclusion should be that Loehle's method overestimates TCR by 24.5% (rather than 14.5%) based on the RCP8.5 forcing data – or by 36.1% based on the more up to date AR5 forcing estimates.

Interestingly, Craig Loehle has noted in the subsequent comments that he was not given the opportunity to respond to Cawley et al before the paper was published.

Update on Jan 13, 2015 by Registered CommenterBishop Hill

In the comments, WB notes that Robert Way posted a series of tweets about Loehle's paper last year. Concatenated, they read:

Last year, this rubbish [i.e. the Loehle paper] was published in an out-of-topic journal that contained the author on its advisory board. Beyond the 4-month timeline from receipt to published, clear statistical errors showed it had not been scrutinized enough during review. Reimplementing their analysis showed a number of methodological flaws and assumptions to the point where a response was necessary. Enter Cawley et al (2015) who show how fundamentally flawed Loehle (2013) was in reality.

Too funny.

Update on Jan 13, 2015 by Registered CommenterBishop Hill

ATTP is doing some hilarious handwaving:

I think that is a genuine error in the Cawley et al. paper, in that they got that Loehle’s method under-estimates TCR, rather than overestimates it (by mutiplying rather than dividing). A silly mistake I imagine since it is trivial, and I don’t doubt all those involved in the paper could easily do this. Of course, what Nic Lewis fails to point out is that the Loehle paper is rubbish and full of completely unjustified assumptions. That it’s already low TCR estimate should actually be lower if he’d done it with more reasonable numbers, just highlights this even more. Would be nice if he focused on the interesting part of the Cawley paper, which I discuss here, rather than nit picking something minor and rather trivial (however, this does seem to be his normal style, so not surprised).

Article originally appeared on (
See website for complete article licensing information.