An exascale computing facility modelled on the organisation of CERN would enable a step-change in quantifying climate change, argue Tim Palmer and Bjorn Stevens.
In the early 1950s, particle accelerators were national-level activities. It soon became obvious that to advance the field further demanded machines beyond the capabilities of single countries. CERN marked a phase transition in this respect, enabling physicists to cooperate around the development of one big facility. Climate science stands to similarly benefit from a change in its topology.
Modern climate models were developed in the 1960s, but there weren’t any clear applications or policy objectives at that time. Today we need hard numbers about how the climate is changing, and an ability to seamlessly link these changes to applications – a planetary information system for assessing hazards, planning food security, aiding global commerce, guiding infrastructural investments, and much more. National centres for climate modelling exist in many countries. But we need a centre “on steroids”: a dedicated exascale computing facility organised on a similar basis to CERN that would allow the necessary leap in realism.
To be computationally manageable, existing climate models solve equations for quantities that are first aggregated over large spatial and temporal scales. This blurs their relationship to physical laws, to phenomena we can measure, and to the impacts of a changing climate on infrastructure. Clouds, for example, are creatures of circulation, particularly vertical air currents. Existing models attempt to infer what these air currents would be given information about much larger scale 2D motion fields. There is a necessary degree of abstraction, which leads to less useful results. We don’t know if air is going up or down an individual mountain, for instance, because we don’t have individual mountains in the model, at best mountain ranges.
In addition to more physical models, we also need a much better quantification of model uncertainty. At present this is estimated by comparing solutions across many low-resolution models, or by perturbing parameters of a given low-resolution model. The particle-physics analogy might be that everyone runs their own low-energy accelerators hoping that coordinated experiments will provide high-energy insights. Concentrating efforts on a few high-resolution climate models, where uncertainty is encoded through stochastic mathematics, is a high-energy effort. It would result in better and more useful models, and open the door to cooperative efforts to systematically explore the structural stability of the climate system and its implications for future climate projections.
Working out climate-science’s version of the Standard Model thus provides the intellectual underpinnings for a “CERN for climate change”. One can and should argue about the exact form such a centre should take, whether it be a single facility or a federation of campuses, and on the relative weight it gives to particular questions. What is important is that it creates a framework for European climate, computer and computational scientists to cooperate, also with application communities, in ways that deliver the maximum benefit for society.
A number of us have been arguing for such a facility for more than a decade. The idea seems to be catching on, less for the eloquence of our arguments, more for the promise of exascale computing. A facility to accelerate climate research in developing and developed countries alike has emerged as a core element of one of 12 briefing documents prepared by the Royal Society in advance of the United Nations Climate Change Conference, COP26, in November. This briefing flanks the European Union’s “Destination Earth” project, which is part of its Green Deal programme – a €1 billion effort over 10 years that envisions the development of improved high-resolution models with better quantified uncertainty. If not anchored in a sustainable organisational concept, however, this risks throwing money to the wind.
Giving a concrete form to such a facility still faces internal hurdles, possibly similar to those faced by CERN in its early days. For example, there are concerns that it will take away funding from existing centres. We believe, and CERN’s own experience shows, that the opposite is more likely true. A “CERN for climate change” would advance the frontiers of the science, freeing researchers to turn their attention to new questions, rather than maintaining old models, and provide an engine for European innovation that extends far beyond climate change.