James Kinter, director of the Center for Ocean-Land-Atmosphere Studies at the Institute of Global Environment and Society, was granted permission to use systems provided by XSEDE - and its predecessor TeraGrid - to run climate models with high spatial resolutions, requiring large amounts of processing power.
Based on these simulations, Kinter
colleagues calculated that the likelihood of extreme drought in the US Great Plains and Florida is set to increase from the average frequency of four out of every 50 years to 20 years of extreme drought out of every 50 years by the end of the 21st century.
Using the XSEDE infrastructure, Kinter
colleagues were able to simulate US climate over long periods at the spatial resolution normally used for producing 10-day weather forecasts today.
"The pattern of increasing probability of extreme drought in our simulations is quite similar to the summer 2012 drought severity map showing 63% of the lower 48 US states in moderate to severe drought," says Kinter
"Our results suggest that, while the 2012 event itself cannot be ascribed to human-induced climate change per se, the severe situation we are experiencing today is likely to become entirely too commonplace in the future."
However, not all climate researchers are as fortunate as Kinter
colleagues to be granted dedicated access to some of the US National Science Foundation's finest high-performance computing equipment.
And, with competition for access to supercomputers fierce, climate scientists are often required to make pragmatic decisions, such as limiting the spatial resolution of their models, so as to save on computing time.
After all, even at moderate resolutions, one could reasonably expect global-scale models to require several months of supercomputing time to produce just a century's worth of climate data.
Speaking at the recent XSEDE12 conference in Chicago, Kinter
discussed the significant effects that changes in spatial resolution of climate models can have on their outcomes.
Working with the Athena supercomputer in 2009/10, he
team were able to run simulations of boreal summer climate at a 7-kilometer resolution over the course of eight summers, whereas researchers had previously only been able to simulate a single week or month at this level of detail.
While running these simulations, the team found that even small changes in spatial resolution could have large impacts on the outcomes of the climate models.
cites research by collaborators that showed how low-resolution models of the Gulf Stream off the east coast of the US put rain associated with the weather pattern in the wrong place, whereas high-resolution models were able to delineate the bands of rain off the east coast with accuracy.
Of course, while ever-faster supercomputers mean that climate models can feasibly be run at increasingly high spatial resolutions, this doesn't mean that a solution to the difficulties of climate modeling is yet in sight.
The research conducted by Kinter
team was pioneering work and required many ad hoc fixes and workarounds to complete the simulations.
For example, the team suffered with issues relating to the memory available per node on Athena and the output from their simulations was capped by the bandwidth limits between both processors and disk and disk and HPSS tape.
According to Kinter
, turning the ad hoc solutions he
team came up with to deal with these issues into systematic, repeatable solutions is the next step which needs to be taken.
However, some of the issues his
team faced were even more difficult to overcome: "We were in a unique situation, because we had dedicated access to Athena 24/7 for six months.
That introduced unique challenges like how to keep the queue loaded with jobs so that the machine never went idle and how to manage the output data.
We generated output at a rate that would fill Athena's disks every 6 days, so we had a fairly serious data management challenge.
The project generated nearly 900TB of data in total and this is now available to researchers around the world hoping to improve the next generation of climate models.
The following list of people collaborated with Kinter
on this research: