Beyond Pattern Scaling: Statistical Emulation and its Implications for ScenarioMIP

One of the crucial aspects of climate policy in the near future is the design of mitigation strategies. However, we can only get information from state of the art climate models at a handful of mitigation scenarios (e.g. the SRES and RCP scenarios). In order to compare alternative strategies and their impacts using models, we need to consider the climate effects of the corresponding emission/concentration pathways for which we don’t have climate model output. Currently this is mainly done by pattern-scaling – multiplying the mean pattern of climate change by the change in the mean of the variable (e.g. temperature or precipitation) over time. A generalised alternative to pattern scaling is statistical emulation.

A statistical emulator is a more sophisticated way of interpolating climates between a limited number of model runs. Although widely used in science for the modelling of computer experiments, the use of statistical emulators in climate science has been limited and mainly used for perturbed physics ensembles. However emulators are perfectly well suited for use with forcing conditions instead of (or as well as) model parameters. Pattern scaling is a special (very simple) case of an emulator.

We will show how by parameterising the forcing functions and building emulators we can predict the climate for any reasonable set of forcings (including overshoot scenarios and other ‘odd’ forcing pathways). We also set out how a ScenarioMIP type experiment would have to be configured to achieve this.