ABSTRACT

The difficulty of dealing with limited computer power is felt more keenly in climate science than in most other disciplines. A key idea allowing calibration in many of these expensive simulators is the idea of replacing the simulator with an emulator, which is a cheap statistical surrogate used in place of the simulator. Climate science presents the double whammy of computationally expensive simulators, and simulator discrepancy that is too large to ignore, but which is not well understood or modelled. History matching, like calibration, seeks to identify regions of the input space that give acceptable matches between simulator output, Csim, and observed data, Cobs. Climate change is driven by changes in CO2 concentration and it is therefore determined by the interplay between anthropogenic emissions and the carbon cycle. The Emulator Filtered Plausibility Constrained (EFPC) parameter set is constrained to simulate a plausible pre-industrial climate, but no constraint was imposed upon the dynamic response to anthropogenic emissions.