HOW CAN WE BUILD BETTER EARTH SYSTEM MODELS?

The most important and interesting variations in Earth's climate have arguably occurred on time-scales of millennia to millions of years, although there have been numerous amazingly rapid major transitions taking only a few decades. There is suggestive evidence that all the major components of the Earth System play important roles in this natural variability. The range of time-scales involved is enormous, and the system has a complex structure, and is therefore difficult to model. Moreover, we need to model spatial variations, not least because we need to distinguish between land, sea and ice. Modelling the fluid processes in both the oceans and the atmosphere at an adequate spatial resolution also introduces serious computational restrictions, which are not easy to overcome.

For the very longest geochemical timescales, box models are useful and not inappropriate. For the interesting range of glacial/inter-glacial timescales, 2 and 2.5 dimensional models have been applied with considerable success. 3D models require modest horizontal resolution (ca 20 by 20) just to represent the configuration of the continents recognisably. Even coarse resolution GCMs (ca 100 by 100) are generally restricted to integration times of a few millennia, with few exceptions. It is suggested that for the future we should concentrate on 3D models of moderate/coarse resolution, suppressing "fast" processes, including eddies, and concentrating our effort on the great intellectual challenge of deriving adequate parameterisations of their effects, using inter alia the results from short integrations of high-resolution GCMs. Several successful instances of this approach exist for the oceans, but the atmosphere seems to be more difficult. Modelling relatively small-scale cryospheric and biogeochemical processes (especially in sediments and soils) within the "fluid framework" thus established remains an interesting challenge.