Cutting-edge computational research to investigate and address the climate change challenge
Giovanni Aloisio
University of Salento & CMCC Foundation, Italy
Modelling the climate system is a big challenge, requiring the simulation of several interacting and complex processes, as well as their analysis at different time and spatial scales. Climate system modelling also requires sophisticated numerical models, due to the inherently non-linear governing equations and huge computational resources are needed to solve billions of individual equations describing the physical processes at different scales. Moreover, one of the main issues in climate science is the quantification of uncertainty and its reduction. This requires increases in model resolution and large-scale ensemble runs to generate accurate statistical information. Access to large computational capabilities for climate modelling is then required to meet the need for higher spatial and temporal resolution, better physical process representation, explicit modelling of more biogeochemical processes, much longer runs and larger ensembles. Clearly, as the resolution increases, extreme data issues also arise, since a larger amount of data needs to be managed at and moved across the different memory hierarchy levels of the computing system. However, the climate modelling community is struggling to exploit the current generation of computing systems efficiently, as most of the climate codes have not been designed to efficiently use the available cores, which are already heading towards tens and hundreds of thousands. Indeed, the achieved efficiencies are currently well below 10% of peak performance for most climate codes. The situation is likely to get worse over the next years, as the memory per core and memory bandwidth per core fall causing inter-processor communication overheads to increase even further. It seems clear that to contend with the challenges posed by current and expected architectures, many of the algorithms and numeric approaches currently in use need re-designing. A co-design approach, involving mathematicians, climate and computational scientists and technology providers, is required to solve these scientific and computational grand challenges.