UB Researchers Use Computer Cluster to Create Model of Contaminant Flow in Groundwater

BUFFALO, N.Y. -- When engineers conduct research on groundwater, the water that flows beneath Earth's surface, they usually think of "large-scale" as one watershed -- an area of land where all of the water on it or under it drains into the same place, such as a lake and its tributaries.

Using a 300-node Dell high-performance computing cluster (HPCC) in the Center for Computational Research (CCR) at the University at Buffalo, Alan Rabideau, Ph.D., and Igor Jankovic, Ph.D., are working to turn that definition on its head.

The goal of the researchers is to create what could be the first groundwater model capable of accurately representing how contaminants flow through groundwater in multiple watersheds at a variety of scales, ranging from the size of individual grains of soil to bodies of water that cover many miles.

"Ultimately, we're interested in problems like climate change and its impact on the Great Lakes," says Rabideau, an associate professor in the Department of Civil, Structural and Environmental Engineering in the UB School of Engineering and Applied Sciences.

In addition to Jankovic, UB assistant professor of civil, structural and environmental engineering, the other members of the team are Douglas Flewelling, Ph.D., UB assistant professor of geography, and Matt Becker, Ph.D., UB assistant professor of geology.

The team is focusing on developing more efficient algorithms so that the resulting models accurately represent groundwater flow at many scales. The project requires the power of a computer cluster that has many, independent processors.

The groundwater research group "graduated" to the 300-node Dell cluster it now uses from a Dell mini-cluster that it used grant money to purchase last year.

"The Dell mini-cluster was unbelievably bulletproof," says Jankovic. "You could not crash that machine."

It's not for lack of trying. Last year, Jankovic's simulations of the behavior of contaminants through groundwater chewed through 280,000 central processing unit hours, the equivalent of 32 years of computing time on a personal computer.

"I use the computer to describe the microscopic movement of contaminants," explains Jankovic, "which then helps us to identify the parameters for the equations in the model. In other words, I simulate the system on a microscopic scale in order to get parameters for simulations on a regional scale," he says.

At the same time, Rabideau studies the chemical end of things, examining and accounting for how differences in individual soil particles influence contaminant transport in groundwater.

"Groundwater systems are fascinating because of the relatively large variations in soil properties that occur over relatively small distances," says Rabideau. "These differences must be accounted for in computer models used to assess the movement and cleanup of pollutants."

In addition to effectively describing processes that occur over multiple scales, the research will require integrating the models with field data and geographic information systems, tasks that Becker and Flewelling are carrying out.

"These challenges cannot be met without the power of multiple, high-speed processors," explains Jankovic.

Existing models of groundwater flow, he adds, were developed decades ago, mostly by the U.S. Geological Survey.

"They have updated these models and added new features, but they were never designed for modern computer architectures," says Rabideau.

Consequently, 21st-century challenges, like studying the complexities of climate change, cannot be tackled adequately using the models.

"If precipitation levels change, for example, as a result of climate change, how is that going to affect levels on the Great Lakes and how will that, in turn, affect groundwater supplies around the lakes?" asks Rabideau.

The exchange between groundwater and the Great Lakes is not well-understood, he added, but it is becoming increasingly important as cities, like Chicago, that border the lakes, increase their reliance on groundwater.

"Traditionally, groundwater models are developed for a particular area in a watershed," explains Rabideau. "The engineers divide the area into a grid, and apply computational tools to do the simulations that are proportional to the size of the area.

"Our approach is fundamentally different. We are developing models not based on the size of a given area, but rather on hydrologic or chemical features that drive the groundwater, such as the number of lakes or changes in soil type."

Such detailed simulations will provide a much more comprehensive view of groundwater, data that will be necessary as demands on groundwater grow and to ensure monitoring of the safety of groundwater supplies that face increasing threats from non-point source pollutants or even terrorist attacks, explained Rabideau.

"We know that contaminants of any kind are incredibly difficult to get out of a groundwater source," he says.

The UB researchers plan to demonstrate their new simulations of groundwater flow on the small scale using Ischua Creek, just south of Buffalo; at the mid-sized scale, using the Trout Lake area in northern Wisconsin, home to hundreds of lakes, and ultimately, the entire Great Lakes watershed, at the large-scale.

The research was initiated by a $1 million Environmental Protection Agency grant awarded to Rabideau in 2000 to do groundwater simulations using supercomputers. Since then, members of the team have been awarded additional grants totaling more than $500,000 by the National Science Foundation.