Reconciling climate sensitivity estimates – part III, or IV?

I mentioned in a recent post that there was a new paper that claims to reconciles climate sensitivity estimates. The basic issue is that observationally-based climate sensitivity estimates have a range of about 1.5^oC to about 3^oC, while climate model estimates tend to range from about 2^oC to about 4.5^oC. However, if you use climate models to estimate something equivalent to an observationally-based climate sensitivity, you get something consistent with these estimates.

However, to the surprise of noone, Nic Lewis has already claimed that their analysis is wrong. His argument appears to be that they made an error (I don’t know if they have) and that the equivalent observationally-based estimates you would get from climate models is higher than they claim. Also Nic Lewis claims that the range for the actual observationally-based estimates is lower than they claimed in the paper. Therefore their claim that these different estimates have been reconciled is wrong.

I don’t know if Nic Lewis has really found an error. That it would appear impossible to publish a paper that reconciles these climate sensitivity estimates without Nic Lewis claiming to have found some kind of fundamental error might suggest caution. However, based on some tweets from Kyle Armour, even if there is an error, it would seem not as obvious as Nic is claiming (not the first time that has happened). Since, however, I don’t have the time to really work out whether he is right, wrong, or somewhere inbetween, I thought I would make some general comments.

• A key point is that there are indications that observationally-based estimates for equilibrium climate sensitivity are biased low. This is essentially because they assume that the feedback response in the future will be the same as in the past (linear). There are lots of reasons why this is unlikely to be the case. As mentioned in this Isaac Held post, we know that some regions warm faster than others. This will manifest itself as warming faster during the final stages of approaching equilibrium, than during the early stages. There are also indications that the pattern of sea surface warming has lead to a cloud feedback response that is smaller (more negative) than what we might expect to be typical. Therefore we should treat observationally-based climate sensitivity estimates with caution; they may well be biased low.
• In Nic Lewis’s criticism he appears to decide that what should be compared are the best estimates, rather than the ranges. He recomputes the model best estimate and claims it is higher than suggested in the paper, and then compares that to the best estimates from observationally-based estimates (which he claims is between 1.6^oC and 2^oC). This itself seems statistically questionable, but there are also potential issues with how some of the observationally-based best estimates are determined. Nic Lewis, for example, uses a Jeffrey’s prior in his analysis. The Jeffrey’s prior potentially allows for climate sensitivity values that are unphysically low and, consequently, could shift the distribution so that the best estimate is biased low (I think he uses the median, but I can’t quite remember). We had a lengthy discussion of this issue in this post.
• What about some ballpark estimates? We’ve already warmed by about 1^oC. If we continue increasing our emissions, we could double atmospheric CO₂ by the middle of the century. The long-term warming trend suggests that we could easily have warmed by 1.5^oC at that stage. Given the thermal inertia of the oceans and estimates for the planetary heat uptake rate, it seems highly unlikely that the TCR-to-ECS ratio could be larger than 0.8. Hence, if the TCR is around 1.5^oC, that would probably mean the ECS is at least 2^oC, which is right in line with the kind of minimum we’d expect, based on our understanding of the various feedback processes.
• There is also an interesting potential irony to all of this. On the one hand, we sometimes see arguments that we can’t trust climate models because the climate system is non-linear and, hence, chaotic and therefore we can’t make predictions about the future. On the other hand, Lewis’s position is essentially that the system is so simple that we should trust linear models over more complex climate models.

Ultimately, I think that the observationally based estimates, some of which have been done by Nic Lewis, are very useful ways of estimating climate sensitivity. However, they are quite simple and are not really inconsistent with the other estimates (ranges from 1^oC to 3^oC compared to other estimates that have ranges from about 2^oC to 4.5^oC). Searching for errors in any attempt to further reconcile the various estimates doesn’t really make the low climate sensitivity values suggested by these observationally-based estimates more credible. To do that would really require showing that they’re consistent with our understanding of physical climatology. I would argue that there are indications that they are not.