Climate change challenges HPC
Scientists spoke about the challenges of climate change modelling and improving simulations to model smaller and smaller systems at this year’s International Supercomputing Conference.
The trio of researchers from around the world agreed that modelling climate change is a tricky business, with simulations producing more data, over massive timeframes and increasing resolutions and that HPC has to try to keep up with the ever-expanding science.
Researchers are trying to understand the Earth as a system, which has a series of subsystems, such as the oceans, atmosphere and biosphere.
Recently, the amount of data has increased enormously which is, in part, due to the rapid rise in computing power that has allowed scientists to use models with higher resolution and perform long-term simulations. The scientists are able to couple models for the subsystems in complex cumulative simulations producing petabytes of output, which is collected in distributed data archives.
And data is also produced by satellites that are monitoring the Earth and producing results in a second huge data stream for climate research.
The Collaborative Climate Community Data and Processing Grid (C3-Grid) links these distributed data archives and provides uniform access to these huge amounts of data.
Speaking at the conference, Professor Wolfgang Hiller, director of the Alfred Wegener Institute for Polar and Marine Research, said: 'The user has to cope with new challenges and has to cope with enormous data production rates and increased complexity of accessing data.'
The C3Grid has recently been updated to make it easier for users to access the data. Hiller added: 'The previous user scenario for C3Grid was very tenuous, but now all the user has to do is sit at their desk and select the data and workflows they need. It is much more transparent.'
And Professor William Collins, department head at the Lawrence Berkeley National Lab, is also trying to cope with increasingly intricate simulations: 'Climate change models are getting increasingly complex and are no longer just predicting the physics of the planet, but incorporating physics, chemistry and biology into the predictions.'
Collins added: 'And including chemical and biogeochemical components increases the demand on the computer power by around five times.'
But it’s not just the models that must include more science; the resolution of these simulations is also increasing, bringing with it another set of problems for the scientists.
Dr Sebastien Masson from the University Pierre and Marie Curie in France, is also modelling climatic change by breaking up the planet into chunks and simulating what happens within each zone to get an overall picture of how the planet's climate is altering.
As the resolution of the simulations has increased over the years, scientists are struggling to decrease the size of the domains any further while maintaining the long time scales that climate modelling uses, because the computing power is simply not sufficient. Masson said: 'The problem gets worse when we increase the resolution, as the time step has to come down and we need to improve the parallel efficiency to keep using the same elapsed times.'
Berkeley’s Professor Collins is also working on decreasing the domain sizes of the Earth’s subsystems and said: 'The resolution has increased by a factor of five from when the first model came out in 1990 to when the last one did in 2005.'
Collins’ has managed to resolve a 10km-sized system, but this requires half a petaflop for sustained performance. To bring the resolution down to 1km, 10,000 petaflops would be needed, according to Collins, but this is not plausible at the moment as he said: 'We do not have machines that are stable enough for long-term predictions.'
But by bringing models down to the 10km resolution, scientists have managed to model previously unsimulated systems. Collins said: 'A 10km resolution allows realistic simulations of coastal currents and we have never been able to capture the biogeochemistry in coastal zones before within climate models.'
'We would like to treat this [the coastal zones] at the organism level, but that is a challenge we will be addressing in a few years time,' he added.