NCAR wraps Earth in a box
Scientists' model predicts a warming planet
Daily Camera, January 30, 2005
Its 30 megabytes of source code are the size of a few digital photos, but this software paints pictures of the future.
Earlier this month, NCAR scientist William Collins was part of the canvas. The National Center for Atmospheric Research scientist stood at the big screen in NCAR's Visualization Lab in Boulder, his face changing hue with the surface of the planet as the 21st century blazed to a conclusion in the span of a minute.
The scenario on screen assumed little change in humanity's fossil-fuel reliance. By 2100, the concentration of atmospheric carbon dioxide, a greenhouse gas, would double to 750 parts per million. The model shows average global temperatures rising 6.75 degrees Fahrenheit.
States. The map, projected in NCAR’s Visualization Lab in Boulder, shows projected changes in global temperatures in the year 2100 based on model runs assuming business-asusual
fossil-fuel consumption. Temperatures rise as high as 18 degrees Fahrenheit in areas near the Earth’s poles. Daily Camera/Sammy Dallal
Some areas -- mostly over oceans, which react slowly to temperature change -- stayed a cool white. Collins pointed to a red swath just above Antarctica, which the model predicts will warm 18 degrees Fahrenheit.
"That'll be enough to melt the sea ice in the area," Collins said.
To derive these images and many others, NCAR 's Community Climate System Model churned in some of the planet's fastest computers for much of 2004. The NCAR model is one of the world's most advanced, capable of mathematically incorporating things as subtle as wind on sea ice and smoke from wood fires into the climate's distant future. Its most recent upgrade -- to version 3 -- was the product of an estimated person-century of effort, with about 80 scientists and engineers pitching in over two years.
From February through November, it ran day and night, generating 20 terabytes of climate data, or enough to fill 500 high-end iPods. It's all input for what will be the highest-profile report on climate change ever written, the Intergovernmental Panel on Climate Change's Fourth Assessment Report, expected in 2007. The NCAR team's contribution will be the most comprehensive among the 15 climate-modeling teams from around the world.
Supercomputers at three U.S. sites and one in Japan -- the Earth Simulator in Yokohama, the world's third-fastest computer -- pushed the simulated planet through a combined 5,000 years of climate evolution in 11 months, said Collins, who chairs the scientific steering committee guiding the NCAR model's development.
First, they did a single "pre-industrial" run. Then they ran with eight different input settings for the 20th century, and the results were compared to actual climate data. Then the model headed into the future with multiple scenarios branching off each of those eight runs -- to 2100, 2200, even 2300.
NCAR scientists won't disclose their findings pending upcoming publication in a major scientific journal. They will say that, despite drastic improvements in the model in recent years, the results through the year 2100 are similar to those generated for the IPCC's last major report in 2001.
More than a dozen modeling teams, including NCAR , contributed to the IPCC's Third Assessment Report. They estimated that the planet would warm between 2.5 and 10.4 degrees Fahrenheit by 2100. They also predicted sea-levels would rise between 3.5 inches and nearly three feet, although such estimates don't take possible melting of polar ice sheets into account.
Modeling the globally complex
The world is full of computer models, from mortgage calculators to flight simulators. Climate models are about as tough as they get.
NCAR's Community Climate System Model seeks to predict the ebbs and flows of heat on Earth. It divides the atmosphere into 26 vertical layers, the oceans into 40 and land into 10, then breaks it all down into about 4 million three-dimensional grid boxes coating Earth.
In each box, the model crunches through Newton's laws, the principles of quantum electrodynamics and countless other nuggets of scientific theory -- plus some mathematical educated guesses -- to determine how light energy from the sun will be trapped, reflected and absorbed on land, in water and in the atmosphere. It considers the complex interactions of volcanic activity, tailpipe emissions, diesel soot, ocean currents, sea ice, winds over mountain tops, and dozens of other factors.
The model then must, at varying intervals, apply the results of each grid point to ones adjacent, and then apply the changing conditions of the atmosphere's grid points to the ones in oceans and on land -- and vice-versa.
It takes mountains of computing power. NCAR's Blue Sky, the world's 39th-fastest supercomputer according to ranking site Top500.org, can crank through about a 3.5 simulated climate-years a day. That is 1,300 times faster than what's happening outdoors, but it still takes a month to get through a century of simulation, Collins said.
Each improvement to the model slows things down. The one running today has twice the resolution of its predecessor; its treatment of the ocean's heat uptake is more accurate; and it simulates the behavior of atmospheric aerosols such as soot, smoke, dust and sea salt much more comprehensively, Collins said.
"It's really exciting to see the level of detail they're putting into a variety of things," said Susan Solomon, a National
Oceanic and Atmospheric Administration scientist in Boulder who leads the working group examining the scientific basis for
climate change in the upcoming IPCC report.
"The NCAR model is certainly one of the premier world models, and their results will certainly be important along with others
in the Fourth Assessment," she said, referring to the 2007 IPCC report.
Clouding the issue
Still, skeptics of the risks of global warming such as S. Fred Singer view climate models' inability to deal accurately with cloud formation as a fatal flaw. Higher temperatures could mean denser cloud cover and foster lower temperatures, Singer argues, and the models can't simulate those processes with confidence.
Collins said Singer is right -- in part. Cloud formation happens on scales of a fraction of an inch; the NCAR model's atmospheric grid boxes are 90 miles on a side. It's too small to model.
In addition, Collins said, the physics of cloud formation is still poorly understood, and can't be accurately modeled until it is.
"So how do we respond?" Collins said. "We could throw up our hands and say, 'You know, we just can't do anything.' That's ludicrous And that's what they'd like us to do, I think."
Instead, he said, climate scientists run not one but 15 or 20 models with wide-ranging approaches to "forcing" cloud cover into the model -- essentially playing God and ordaining clouds here, clear skies there. It appears to make much difference.
"They all predict the climate's going to warm," Collins said.
Advances in computing power may one day eliminate the guesswork.
Leo Donner, a scientist with NOAA's Geophysical Fluid Dynamics Laboratory in Princeton, N.J., said faster machines have allowed researchers to start linking the best cloud-system models with global climate models for the first time.
"Depending on just how fast computing advances, we may even be able to experiment with global models that actually resolve
clouds," Donner said in an e-mail.
© 2008 Todd Neff