August 02, 2012
Numerical weather prediction was one of the original computing problems. When the ENIAC, the first electronic general-purpose computer, came online in 1947, simulations of the atmosphere (along with missile trajectories) was one of the first problems scientists ran on the system.
James Kinter, director of the Center for Ocean-Land-Atmosphere Studies at the Institute of Global Environment and Society, presented this historical tidbit on the second morning of the recent XSEDE12 conference in Chicago. He then showcased the latest advances in climate and weather modeling enabled by the Extreme Science and Engineering Discovery Environment (XSEDE), the National Science Foundation (NSF)-supported cyberinfrastructure for open science.
His talk, “Benefits and Challenges of High Spatial Resolution Climate Models,” included the results of simulations of climate runs between 2008 and 2011 on TeraGrid and XSEDE systems (TeraGrid was the predecessor to XSEDE).
The presentation covered three major research projects funded by the NSF: (1) Project Athena – Resolving Mesoscales in the Atmosphere; (2) PetaApps Team – Resolving Ocean Eddies; and (3) CMMAP – Super-Parameterization and Resolving Clouds, a project led by David Randall at Colorado State University. Cumulatively, these projects, each of which involves dozens of researchers internationally, show the ability of simulations and scientific visualization to depict our warming Earth on a regional scale with uncanny accuracy.
“You might think there’s a debate about climate change,” Kinter said. “But in my community, we’ve gotten past the point of it being a debate. However, our climate models are not perfect.” Climate change deniers leap on these imperfections to challenge whether we can trust the models. “To answer this question, we have to prove the case,” he said.
In the last 50 years, the field of climate and weather modeling has taken advantage of the million-fold increase in computing power to make three improvements to the codes that mimic the atmosphere.
According to Kinter, scientists have improved our understanding of the physical processes involved in atmospheric modeling and incorporated these insights into the evolving codes. They have developed better data assimilation methods to incorporate information from satellites, Doppler radar and ocean monitoring sensors into their models. And they have increased spatial resolution, or the amount of fine-grained detail, that can be included in the simulations.
There is evidence that this last step — enhanced spatial resolution — can not only improve climate model fidelity, but also change our understanding of climate dynamics both qualitatively and quantitatively.
The big question, though, is: “What’s the bang for the buck when you start looking at high resolution?” To test this, Kinter and his colleagues simulated a variety of climate scenarios at resolutions ranging from 7 kilometers (the most fine-grained) to 125 kilometers (the most coarse-grained).
To accomplish this massive computing feat, Kinter’s team was granted a special allocation of computing time on the Athena supercomputer at the National Institute for Computational Sciences (NICS) in 2009 and 2010. For six months, the entire 18,048-core system was at the disposal of the team. Based on those runs and follow-ups on other high performance computing systems, his group has published more than a half dozen publications that run the gamut from the dynamics of tropical storm and cyclone formation to global and regional rainfall forecasts.
Among the results he presented at the conference were simulations that represented boreal summer climatology at 7-kilometer resolution over the course of eight summers. Previously researchers had only been able to simulate a single week or month at this level of detail.
Animation of boreal summer 2009 simulation at 7 km resolution using the NICAM model from JAMSTEC and University of Tokyo.
Earlier simulations produced by many groups around the world showed trends of modeled surface temperature change over the last century that have a statistically significant separation at the global and large continental scale between simulations that include the human influence on climate (increasing greenhouse gases and aerosols) and those that don’t. This was “the smoking gun of whether humans are responsible for the rise in temperature,” Kinter said.
However, the trends at regional scale are not as discernible. Is that because the trends are not there or because the models lack the acuity to see them? Kinter and his colleagues’ investigations of high spatial resolution shed light on this question.
Other simulations explored the probability of extreme drought in the Midwest, Europe and elsewhere in the future. By his estimates, the Midwest will experience the levels of extreme drought it is currently experiencing in 20 years out of every 50 — a four-fold increase. “This drought will be the norm at the end of the 21st century,” Kinter said, “according to these simulations.”
He also presented a number of key examples where increases in model resolution impacted the clarity and content of results. For instance, he cited research by collaborators that showed how low-resolution models of the East Coast Gulf Stream put rain associated with the weather pattern in the wrong place, whereas high-resolution models delineate the bands of rain off the East Coast with accuracy.
After outlining the advantages of higher-resolution models, Kinter elaborated on the challenges that such a change generates. Biases in the models, the parameterization of small time and spatial scale effects (like clouds), and the coupling of global climate models with cloud resolving models, are all difficult, but not impossible, to overcome. However, the primary challenge that Kinter’s group and the community are dealing with is the “exaflood of data” produced by high-resolution and highly complex coupled models.
For Project Athena, the total data volume generated and now resident at NICS is 1.2 petabytes. However, the total data volume on spinning disk at the Center for Ocean-Land-Atmosphere Studies for Project Athena is capped at 50 terabytes. This creates difficulties.
Running on TeraGrid systems at large-scale for the first time with so much data, “everything broke,” Kinter said. He and his colleagues had to find ad hoc solutions to complete the simulations. The next step, he said, is to take those ad hoc solutions and use them to develop systematic, repeatable solutions.
Put another way: to deal with the exaflood, the community needs to progress from Noah’s Ark to a professional shipping industry. “We need exaflood insurance,” Kinter concluded. “That’s what we’re calling on the XSEDE team to help us with.”
The following contributed to the work described in this article: Deepthi Achutavarier, Jennifer Adams, Eric Altshuler, Troy Baer, Cecilia Bitz, Frank Bryan, Ben Cash, William Collins, John Dennis, Paul Dirmeyer, Matt Ezell, Christian Halloy, Mats Hamrud, Nathan Hearn, Bohua Huang, Emilia Jin, Dwayne John, Pete Johnsen, Thomas Jung, Ben Kirtman, Chihiro Kodama, Richard Loft, Bruce Loftis, Julia Manganello, Larry Marx, Martin Miller, Per Nyberg, Tim Palmer, David Randall and the CMMAP Team, Clem Rousset, Masaki Satoh, Ben Shaw, Leo Siqueira, Cristiana Stan, Robert Tomas, Hirofumi Tomita, Peter Towers and Mariana Vertenstein, Tom Wakefield, Nils Wedi, Kwai Wong, and Yohei Yamada.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?