Sir

By using the distributed power of personal computers around the planet, D. A. Stainforth and colleagues (Nature 433, 403–406; 2005) have quantified uncertainty in forecasts of global warming resulting from a doubling of CO2. If warming occurred at the upper end of the predicted range, the effects for humankind would be utterly catastrophic.

Such forecast uncertainty arises, in large part, from the way cloud systems are represented in existing global-climate models. Because of limitations in computer power, the underlying laws of physics cannot be used directly to simulate such systems; instead they are represented by approximate formulae with uncertain parameters. In the paper by Stainforth and colleagues, an ensemble of many thousand individual global-warming forecasts was analysed. The forecasts were all made with the same climate model. However, the parameter values varied from one forecast to another — the intra-ensemble variation in these values being consistent with their inherent uncertainty.

Global-climate models are now being developed in which one of the most important types of cloud system, ‘organized deep convection’, is computed explicitly. Such convective systems, with horizontal scales of tens of kilometres, play a key role in climate: they cool the Earth's surface, they transport water from the surface into parts of the troposphere where they can contribute to the greenhouse effect, and their kinetic energy influences global-scale climate circulations. Unfortunately, even on the most powerful computers, climate simulations in such models take nearly as long as real time. Such high-resolution climate models cannot be run efficiently using distributed computing technology.

To reduce uncertainty in estimates of global warming, the climate-modelling community requires substantially more powerful computational resources than are currently available, so that more of the climate system can be simulated directly from the known laws of physics.

In view of the seriousness of the global-warming problem, plans for developing such a facility for climate prediction should occur at the international level, so that national resources can be pooled and scientific collaborations enhanced. Within Europe, the European Union could play a leading role.