Predicting the future, shrinking to atomic dimensions, traveling back to the birth of the universe: these seemingly impossible feats are all in a days work for a particular breed of scientist called computer modelers. When traditional research is too expensive, dangerous or time-consuming to physically conduct, a computer model can stand in, examining everything from a particular policys effect on the environment to the many possible protein structures resulting from a single DNA sequence. The process of model creation can be excruciatingly slow, however, taking many years.

To address this problem, the U.S. Department of Energy (DOE) has created a new program called Scientific Discovery Through Advanced Computing (SciDAC), which encourages and funds close collaborations between the people who program the computer codes and those who apply and tweak the resulting software. A five-year program, SciDAC already funds 51 projects with a total of $57 million per year.

A Change for the Better

One of the research fields SciDAC has particularly prioritized is climate change. Agriculture, ecosystems and ocean levels are all inextricably linked to the atmosphere--and understanding these processes is obviously critical to studies of climate change and the formation of public policies that are shaped by those studies. Pre-SciDAC, however, the lack of strong bonds among researchers in such disparate areas had made collaboration difficult, hampering and slowing the development and refinement of models. "The primary benefits of SciDAC fall into two categories," explains Dave Bader, acting director. "First, computer simulation will be available as a research tool for a much larger community of researchers than is currently possible. Scientists will no longer be forced to develop their own models to use simulations as a research tool. Second, SciDAC will demonstrate that computer simulation can produce breakthrough basic scientific discoveries."

After two years of preliminary planning and design, SciDAC formally got under way in January 2002. A SciDAC-sponsored conference in Reston, Va., brought applied mathematicians and computational scientists--the code writers and hardware developers--together with "applications" people--the folks who continually use and improve computer models. "The two groups took turns setting up their posters," says participant Dave Randall, professor of atmospheric science at Colorado State University. "While the applications posters were on display, the mathematics and computational scientists people made the rounds looking for matches and vice versa."

After the meeting, 13 interdisciplinary climate groups formed with members from 20 national laboratories and universities. Though each scientist had worked independently for years with some outside partnership, none had ever before collaborated so closely. To further enhance the collaboration, SciDAC has also launched a new program called Science Grid. It links researchers and research groups electronically and allows them to share information, manage gigantic data sets, share computing resources, and communicate with each other more quickly and easily in a secure environment.

Improving Models

Randall heads one research group that's benefiting from SciDAC's funding and encouragement. His team seeks to address a critical problem in modeling: coupling, or bringing together disparate models to create a more complete picture of an environmental system. For example, modeling researchers must bolt together the vital elements of atmosphere and ocean to study processes like the water cycle or heat exchange. Because of the models' divergent structures, however, making them work together is as challenging as plugging an American appliance into a European socket: it can be done, but it requires a lot of effort and forcing--and, in the case of the model, serious computing power and processing time. Randall's team is addressing this problem by creating an ocean model constructed with a grid identical to that of the atmosphere model. "Using atmosphere and ocean grids that have the same shape is in a sense an obvious thing to do, but it is not being done in all models," Randall explains. SciDAC team members are helping to create faster, more efficient codes and hardware for Randall's model, which should speed its completion.

Another technique being pursued under the auspices of SciDAC better reflects the fluidity of atmospheric systems, long a challenge for modelers. Atmospheric models also tend to represent the atmosphere as a stiff, unmoving pile of flat layers--quite unlike reality. To simulate air moving up or down in the atmosphere, models required immense quantities of processing time and difficult codes that were time-consuming to write. Randall, with others, is applying a new technique that allows layers to bend and wave in a realistic fashion, like a flag flapping in the breeze. "Using layers that drift with the air or water--the jargon is quasi-Lagrangian vertical coordinates--is an idea that has been around for a long time--at least 40 years, and arguably even longer," Randall says. "But there were various technical problems that had to be overcome before the idea was ready for practical applications. These problems have been overcome gradually, over a period of decades."

The Search for Resolution

In addition to the atmosphere, models must also include other key earthly elements, such as the ocean, land masses and even sea ice. Each of these components requires the contribution of experts, and Bob Malone of Los Alamos National Laboratory is one such specialist. Working on the oceanic arm of the Community Climate System Model, an interdisciplinary simulation, Malone is striving to increase model resolution from recognizing only uniform water bodies surrounding six or seven blobs of land to distinguishing tiny islands and finger-thin inlets.

William Gutowski, a professor of atmospheric science at Iowa State University, however, has a different approach for the resolution issue. Most models produce output maps at a uniform resolution, requiring a great increase in computing power whenever a researcher demands finer detail. Instead, Gutowski uses technology analogous to holding a magnifying lens over particular areas of the map. "The issue here is that there is always a trade-off in climate modeling between high resolution and acceptable computational cost--primarily time needed for simulation," Gutowski explains. "What we are trying to do is increase resolution only where it appears most necessary." Gutowski and his colleague Joe Prusa, a professor of mechanical engineering at Iowa State, are not the first to do this. But they are among the first to allow the magnifying lens to move with the phenomena that is being modeled--a process dubbed "dynamic grid stretching." "Even that is not really new," Gutowski concedes. "Dynamic grid stretching has been under consideration in some areas of engineering and was also explored for short-term atmospheric simulation at one time. Like most science, we are building on prior work."

Even with the advances in development, the science of improving computer models can sometimes feel like the science of just muddling through. "There can be no recipe for a perfect model," notes Robert Dickinson, a climate modeler at the Georgia Tech School of Earth and Atmospheric Sciences. "The problem is that nature is infinitely complex, so any model has to simplify. Our judgment as to what to simplify includes a certain amount of guessing. We progress by continuing to add more features as we understand the system better and collectively can better see what is important." Gutowski adds, "The problem is that the earth's climate system involves myriad nonlinear processes that are interlinked. One would have to know the exact state of every particle in the earth's climate system for at least one time, which quantum mechanics tells us we cannot know. Additionally, one tiny error in one process will almost invariably amplify and spread into other processes, and so eliminate perfection."

Despite all this, the modelers have hope. "The errors are not so large, it appears, as to prevent approximate representations," Gutowski says. "That is what climate scientists are always working on--improving the approximation."

Rachael Moeller is based in Massachusetts. She earned her master's degree in environmental studies at Brown University.