Since I’ve discussed climate sensitivity here quite a lot, I thought I would highlight a recent paper, by Knutti & Rugenstein, called feedbacks, climate sensitivity and the limits of linear models. It’s a very nice, and readable, summary and – as the title suggests – focuses somewhat on the linear models that are a favourite of Nic Lewis, for example. The basic message, as the abstract says, is
Our results suggest that the state- and forcing-dependency of feedbacks are probably not appreciated enough, and not considered appropriately in many studies. A non-constant feedback parameter likely explains some of the differences in estimates of equilibrium climate sensitivity from different methods and types of data. Clarifying the value and applicability of the linear forcing feedback framework and a better quantification of feedbacks on various timescales and spatial scales remains a high priority in order to better understand past and predict future changes in the climate system.
It also says:
…it becomes clear that ECS and TCR are rather limited characterizations of a much larger and interactive system. Other feedbacks such as vegetation, chemistry or land ice are now included in some climate models as their relevance is better understood. Some feedbacks operate on very long timescales that are determined by the internal dynamics of the system, and their response is not proportional to temperature.
which reminded me of this Michael Tobis comment, where he essentially argues that focusing on a simple metric, like climate sensitivity, ignores a great deal of important complexity. Climate sensitivity might give us some broad brush idea of the magnitude of the change, but it tells us little of what will happen where we live, which will – of course – differ from place to place.
The interesting issue at the moment, however, is that different estimates for climate sensitivity can produce quite different results. For example:
… some but not all recent studies on the twentieth-century warming find rather low ECS values (median at or less than 2°C) [17–19,21]. Climate models show a large spread in ECS, with the spread half as big as the actual value. The highest uncertainty can be attributed to the cloud feedbacks (traceable to certain cloud types and regions), and the lapse rate feedback [50–53]. But all comprehensive climate models indicate sensitivities above 2°C, and those that simulate the present-day climate best [54–57] even point to a best estimate of ECS in the range of 3–4.5°C.
The paper then goes on to discuss the basic linear model, largely represented as
where 
it has been suggested that the non-constancy in the global
is caused by the evolving spatial surface temperature pattern, which (through
) enhances certain local feedbacks at different times [62]. Further, it has been shown that the evolving sea surface temperature pattern alone could explain the time or state dependency of
![\lambda]()
Anyway, I’ve actually said more than I meant to. The paper itself is very accessible, so I would recommend those who are interested in this, and who would like to know more, to go ahead and read it.