A New Promising Approach of Pattern Scaling

Abstract:

Running large ensembles of fully coupled climate models is computationally expensive. This is exactly where climate emulators – among which pattern scaling is perhaps the most popular one – come into play.
The traditional pattern scaling technique assumes that a normalized fixed spatial pattern of a certain variable of interest can be scaled by the trajectory of the global mean temperature change.
Instead of scaling a pattern at a certain period of time from a higher/lower emission scenario, we propose to consider a scaling in time. With this approach, one needs to find a time period with matching global mean surface temperature on an already existing emission pathway. We expect to find the following advantages as compared to the traditional technique:
Consistency between variables (e.g. physical relationship between temperature and precipitation) and correlation in space and time is preserved. Variability does not have to be scaled in magnitude, and is consistent with the global temperature change. And finally, physical limits are preserved (e.g. sea ice loss) to the degree that they depend on global temperature.

It is currently unknown whether different forcing agents (aerosols, land use change) which differ in magnitude across scenarios might cause a problem with the proposed approach, but the same problems arise with scaling patterns between scenarios at a fixed time.
We first test this new approach with one model (CESM) and then with a multi-model ensemble (CMIP5). Due to the interdependency in a multi-model ensemble, we test the sensitivity of our results on the manner the individual models are weighted. In a last step, we aim to consider the effect of forcing, variables (such as climate extreme indices) and model resolution on the viability of the traditional and proposed pattern scaling approaches.