A new way of quantifying GCM water vapour feedback.

The water vapour feedback probably makes the largest contribution to climate sensitivity, and the second-largest contribution to its uncertainty, in the sense of disagreement between General Circulation Models (GCMs, the most physically detailed models of climate we have). Yet there has been no quantification of it which allows these differences to be attributed physically with the aim of constraining the true value. This paper develops a new breakdown of the non-cloud LW (longwave) response to climate change, which avoids the problems of the conventional breakdown, and applies it to a set of 4 GCMs. The basic physical differences are that temperature is used as the vertical coordinate, and relative humidity as the humidity variable. In this framework the different GCMs' feedbacks look more alike, consistent with our understanding that their water vapour responses are physically very similar. Also, in the global mean all the feedback components have the same sign, allowing us to conveniently attribute the overall response fractionally (e.g. about 60% from the 'partly-Simpsonian' component). The systematic cancellation between different feedback components in the conventional breakdown is lost, so now a difference in a feedback component actually contributes to a difference in climate sensitivity, and the differences between these GCMs in the non-cloud LW part of this can be traced to differences in formulation, mean climate and climate change response. Physical effects such as those due to variations in the formulation of LW radiative transfer become visible. Differences in the distribution of warming no longer dominate comparison of GCMs. The largest component depends locally only on the GCM's mean climate, so it can in principle be calculated for the real world and validated. However, components dependent on the climate change response probably account for most of the variation between GCMs. The effect of simply changing the humidity variable in the conventional breakdown is also examined. It gives some of this improvement-the loss of the cancellations that leave the conventional breakdown of no use to understand differences between GCMs' climate sensitivities-but not the link to mean climate. [ABSTRACT FROM AUTHOR]