For 30 years the IPCC said that ECS is most likely around 3°C, plus or minus 1.5°C. As we explained in our video on ECS in 2007 they briefly moved the bottom end up to 2°C, but the evidence kept rolling in that ECS values below 2°C are quite likely, so in the 2013 report they moved it back down to 1.5°C. But luckily for the IPCC, just in time for their 6th Assessment Report a new paper by Australian climatologist Steven Sherwood and his coauthors appeared in 2020 applying a fancy new method to combine not just recent climate data but model and paleoclimate data, yielding an ECS best estimate of 3.2°C and a likely range between 2.3°C and 4.7°C. On that basis the IPCC declared the ECS range had to be above 2.5°C and anything lower was très impossible. But when Nicholas Lewis looked closely, he found the new study had used some faulty math and outdated data. In a 2022 paper he fixed the problems yielding an ECS of about 2.2°C, and perhaps even lower. Once again the IPCC jumped on a study that told them what they wanted to hear, without checking the math. And once again they got caught.

The new study employed a method called Bayesian statistics. The main idea of Bayesian methods, in contrast to regular or “classical” methods, is that you take account of your prior belief when doing the calculation. Suppose you want to estimate the average age of children in a school and your sample of 5 observations is 9, 9, 10, 11 and 11. Estimating the average is simple: add up to get the total of 50, then divide by 5 to get 10. Bayesian analysis however says, suppose you have a prior belief that the average age is 7. In light of the data sample what is your estimate of the average age? Whether the answer is something other than 10 comes down to how much weight you put on the sample data and how much on your prior belief.

Sherwood and crew combined three types of data, model-generated, paleoclimate and modern instrumental measures, along with Bayesian “priors” (i.e. their assumptions about ECS), to argue that the best estimate was even higher than previously thought and the lowest likely value was 2.3°C. Lewis argued that the Sherwood study did the math incorrectly and used outdated data.

The first problem was they did the Bayesian step in a way that biased the answer, basically putting too much weight on an assumed answer and not letting the data drive the result. Also they used input data that had already been updated in the IPCC report itself. Putting everything together Lewis concluded the likely range was 1.6°C to 3.2°C, with a best estimate of 2.2°C. And he pointed out that far from it being impossible to have an ECS under 2.5°C as the IPCC had claimed, in fact it is more likely than not, with a 36% chance it’s below 2°C and a 38% change it’s between 2° and 2.5°, with only a 26% chance it’s above 2.5C°.

Interestingly, Lewis also noted that when incorporating the “modern” data, Sherwood et al. used surface thermometer records from before 1870, even though that part of the record is very uncertain. If Lewis dropped the pre-1868 portion, the ECS best estimate fell further to 1.8°C. So we’re back once again to an ECS best estimate below 2°C, meaning no cause for panic.