This is going to be a rather lazy post, as I’ve been at the beach all day with the kids and don’t have much energy. I was just interested in whether or not anyone had had a chance to look at Nic Lewis’s latest paper (Implications of recent multimodel attribution studies) which is due to appear in Climate Dynamics. It seems to be mainly an update on his 2014 paper in which he uses an Objective Bayesian approach to infer climate sensitivity.
I tried working through Nic Lewis’s latest paper, but there was lots of discussion about Bayesian analysis, and as much as I’d like to claim that I understand Bayesian statistics, I don’t really – well, not at a level that allows me to work through Nic Lewis’s paper. The basic results, however, are illustrated below.
So, ECS and TCR ranges that are not unreasonable, but somewhat lower than many other estimates. I’ll make a couple of quick comments.
- As I understand it this is a basic energy balance approach and so cannot capture all the complexities of our climate. For example, it assumes feedbacks are linear. To be fair, Nic Lewis does discuss some of this at the end of his paper.
- The IPCC ECS likely range is 1.5K – 4.5K. At the recent Ringberg meeting, it appeared that many regard the ECS range as probably being between 2K and 3.5K. Even though Nic Lewis’s results are reasonable, they still seem to be at odds with what most other experts regard as likely. His results suggest that the ECS is more likely to be below 2K than above, while many others seem to regard it as more likely to be above 2K, than below.
- This is where I might potentially embarass myself, but my understanding of one of the strengths of Bayesian statistics is that you can incorporate prior knowledge. Nic Lewis uses an Objective Bayesian approach which – as I understand it – means that he regards his prior assumption as objective. However, our basic prior knowledge is that ECS is probably above 2K. That his method produces a result that suggests the ECS is probably below 2K, might suggest that some kind of physically motivated prior might be preferable to one that is regarded as objective. This seems to be roughly what James Annan is saying here.
Anyway, that’s all I was going to say. If anyone else has any thoughts, or has actually worked through the paper and understands it better than I do, feel free to present them through the comments.