The third version of NCAR's flagship climate model is Linux-friendly and packed with improvements.
by Bob Henson, David Hosansky, and Lynda Lester
University scientists—and everyone else interested in divining the future of global climate—have a powerful new tool at their disposal. Version 3 of NCAR's Community Climate System Model debuted on 23 June. Freely available to users, CCSM3 offers new versatility and sophistication in its depiction of air, land, oceans, and sea ice.
"We've gone through a very intense period of development," says William Collins (pictured at right), chair of the CCSM scientific steering committee. Shortly after successfully releasing the model, the CCSM team introduced Version 3 at their annual users' meeting, held in July in Santa Fe, New Mexico. More than 100 people took part in tutorials on how to operate the new model.
For nearly a decade, NCAR's Climate and Global Dynamics Division has led the development of CCSM as a state-of-the-art system tailored to researchers in academia. The development has been supported by funding from and scientific collaboration with NSF, the U.S. Department of Energy, NASA, and NOAA. "The university community has been very actively engaged in its development, so they're not simply end users," says Collins. "We hope they use it as a tool both for doing their own research and for working with graduate students and postdocs."
To help bolster the new model's credibility and usefulness, NCAR carried out a number of multi-century control runs. The data are available on the project's Web site. Preliminary results indicate that the new version yields greater surface warming than the last version when carbon dioxide is increased to twice its present-day value. Several scenarios for emissions suggest that the atmospheric concentration of CO2 could double by 2100.
Researchers have yet to pin down exactly what is making the CCSM3 more sensitive to CO2, according to Collins. However, the model performed impressively in simulating the variations in the observed reconstructions of 20th century temperature. That, along with other test results, makes Collins and his peers confident that the new model represents a substantial ramp-up in accuracy from CCSM2.
Pick your resolution
In a sense, CCSM3 is three models in one, each optimized for a specific resolution and purpose. The most complex of the three was built with the Intergovernmental Panel on Climate Change (IPCC) in mind. NCAR scientists and university collaborators are running many of the key simulations for the IPCC's next major assessment, which is due in 2007.
The high-res CCSM3 features four times the number of data points as CCSM2 for its land and atmosphere components. To simulate a single day of global climate, the high-res model carries out some 3 trillion calculations. According to Collins, the new version captures such features as continental land temperatures and upper-atmospheric temperatures far more accurately than its predecessor.
The new model is available in two other resolutions:
- Low-res: Suitable for quick experiments to explore new model physics or to perform very long-range simulations of paleoclimate and biogeochemistry.
Intermediate-res: Designed as the workhorse for model development at NCAR and universities, as well as for paleoclimate studies over somewhat shorter intervals.
CCSM meets Linux
With the release of its third version, CCSM has joined the world of Linux users. Each of the three CCSM3 variants above (low, medium, and high resolution) is offered in a version that be can be run on Linux, the open-source operating system that has become increasingly popular on university campuses.
"Our goal was to make it easier to move CCSM3 onto systems that universities can afford," says Collins.
The Linux versions of CCSM3 got a new home at NCAR on 12 July with the arrival of Lightning, a large-scale, high-performance Linux cluster manufactured by IBM. Benchmark tests using CCSM's atmosphere and ocean components showed Lightning to be 30% to 40% faster per processor than Blue Sky, the larger IBM cluster used since 2001 for much of NCAR's climate modeling.
Now being tested by friendly users, Lightning should become available to the broader community sometime this fall. "We expect this system to be a big hit, but we must first prove it to ourselves and our users," says Al Kellie, director of NCAR's Scientific Computing Division.
Testing and collaboration
In order to expedite the IPCC work, the new CCSM has been running IPCC-related simulations this year at several DOE labs and on Japan's Earth Simulator, the world's fastest computer. "We're using the Earth Simulator to build larger ensembles of model runs and to conduct simulations of longer periods than we could do here," says NCAR's Frank Bryan, who began using Japan's flagship computer for ocean simulations last year.
Once the IPCC experiments are wrapped up later this year, CGD researchers will use the Earth Simulator through 2006 to run extremely high-resolution experiments (up to ten times the current resolution), using the atmosphere and ocean components of CCSM. Using computer time funded by the Japanese government, these studies will gauge global climate and help determine whether climate change could alter typhoon patterns in the Pacific Ocean or spur rising sea levels that could affect Japan's coastal communities.
Meanwhile, a number of university scientists are helping NCAR put CCSM3 through its initial paces. At the University of Washington, Cecilia Bitz has been using CCSM to investigate the behavior of sea ice, including a puzzling seasonal aspect. The interaction between sea ice and sunlight occurs in the warm season, and most sea ice reduction over the last century has been during the summertime, yet polar warming is almost exclusively a wintertime phenomenon, Bitz notes. "These apparent paradoxes point toward the importance of simulating sea ice thickness. CCSM3 allows us to simulate such phenomena and others with the best fidelity of any climate model to date. As a result, we're learning more about polar amplification and climate sensitivity and variability."
Much of the improvement in CCSM3 is in the model's foundation for follow-up work, such as in biogeochemistry and land-atmosphere interactions. "The model development never ends," says Collins.
Former NCAR scientist Robert Dickinson (Georgia Institute of Technology) is a longtime land-surface modeler who looks forward to carrying out some of the land-atmosphere simulations made possible by CCSM3."We now have what is in many ways a much more accurate description of how land behaves as part of the climate system, on the scales on which it varies," says Dickinson. "This makes the model a very attractive tool for university scientists."
NCAR scientists are now testing a version of CCSM3 with detailed treatment of ocean-land-atmosphere interaction, including chemical reactions and a range of major airborne particles, or aerosols. This includes "the natural ones, like sea salt and dust," says Collins, along with aerosols whose concentrations are strongly affected by human activities, such as soot from forest fires or factories. The quest for resolution continues, especially in the realm of clouds and convection. Cloud particles form on scales of microns (0.00004 inches), while cloud formation is now simulated in global models on scales closer to 100 km (60 mi). "So there are 11 orders of magnitude separating us from the fundamental phenomena. What we're trying to do is start bridging that gap," says Collins.
Of course, each improvement in a component model makes it more challenging to produce full interactivity in the overall model. That task promises to keep Collins and his colleagues busy for model generations to come.
"We're building a railroad from the east to west coast," he says, "and we haven't yet driven the golden spike."
|What's new in CCSM 3.0
- New treatments of cloud and ice-phase processes
- Improved representation of the interactions among water vapor, solar radiation, and terrestrial thermal radiation
- New treatment of the effects of aerosols, including prognostic sulfate, on the reflection and absorption of solar radiation
- New dynamical frameworks suitable for modeling atmospheric chemistry
- Improved performance and scalability on parallel supercomputers
- Faster multi-way communication among the component models
- New communications infrastructure
- New methods to enable simulation of the terrestrial carbon cycle
- New methods to enable simulation of dynamic vegetation
- Improvements in land-surface physics to reduce temperature biases
- New load-balancing implementation for substantial performance improvement
- Improvements to the representation of the ocean mixed layer
- Variable penetration of solar heating into the upper ocean
- New infrastructure for studying vertical mixing in the ocean
- New advanced sea ice rheology (depiction of deformation and flow)
- Explicit ice-thickness distribution physics
- Explicit treatment of brine pockets (regions of enhanced salt content)
- Improved scheme for horizontal advection of sea ice
- New portability for vector and Linux supercomputers
- New, easy-to-use methods to run IPCC climate-change experiments
- Flexibility to simulate climate over a wide range of spatial resolutions with greater fidelity
- New built-in test facilities suitable for validating installation and verifying some types of model changes
Community Climate System Model
Arrival of Linux supercomputer
(SCD News, 15 July 2004)
Tapping the Earth Simulator
(Staff Notes Monthly, April 2004)