Environmental Audit CommitteeWritten evidence submitted by Andrew Montford

In her evidence to the committee on 12 June 2013, Professor Julia Slingo answered a question from Dr Offord about climate sensitivity as follows:

It’s also really important that we do use observations in some ways to check if our climate models are within sensible bounds. It is still the case that even with the latest updates to what we think what the transient climate response is or the equilibrium climate sensitivity using the last decade of observations, the models—certainly the models we use—and many of the leading models around the world are still within the range of those estimates.

This is misleading. The graph below shows probability density functions for recently published instrumental-observation constrained estimates of equilibrium climate sensitivity (ECS), and the only purely observational study for which an estimated PDF for ECS was included in the last IPCC assessment report, all peer reviewed. Studies including the last decade of observations are in green. Those relying on slightly earlier data, but which use objectively valid methods, are in blue.1 The grey band represents the range of model estimates. It is obvious that there is little overlap between the observationally constrained estimates and the models, the former suggesting a value of climate sensitivity that is considerably less alarming.

Recent observationally constrained estimates of climate sensitivity—Lindzen and Choi 2011, Forster and Gregory 2006, Aldrin et al 2012, Lewis 2013, Masters 2013, and two estimates from Otto et al 2013 (using data from 2000–09 and from 1970–2009). The Lindzen and Choi PDF was not shown in that paper but was calculated from its results by Lewis.

The only way Professor Slingo could conceivably make the statement she did in relation to equilibrium climate sensitivity would be to include studies which used a Bayesian prior that biased their results to high ECS values. That includes studies based on using a uniform Bayesian prior distribution for ECS, an approach that is known to give a warm bias and has been shown in the scientific literature to be invalid.2 ^, 3It also includes studies using a non-uniform Bayesian prior distribution that substantially downweights low ECS values before any observational evidence is introduced. Neither approach would be supported by any reputable statistician.

In relation to the models used by the Met Office, its website states that the HadGEM2-ES model in particular is used for policy advice.4 The HadGEM2-ES model has an ECS of 4.6°C and a transient climate response (TCR) of 2.5°C, the highest TCR and second highest ECS among the 23 CMIP5 models analysed in Forster et al., 2013.5 By comparison, the best observationally constrained estimate of TCR from the recent Otto et al paper was 0.9–2.0°C.

I would urge the committee:

to write to Professor Slingo, asking her to rebut the points made above, stating what published studies she relied on for her assertion regarding the models used by the Met Office and confirming what models she referred to; and

to take evidence from a reputable statistician with expertise in objective Bayesian statistical inference.

18 June 2013