Environmental Modeling

Controlling air pollution has proven to be a more difficult and costly problem than expected. In spite of the $40 billion dollars spent each year, still about 1/3 of the population of the U.S. lives in areas out of compliance with the National Ambient Air Quality Standards set to protect health. It is now recognized one reason greater improvement in air quality was not achieved was that we did not fully understand the critical factors leading to the formation of many pollutants, such as ozone, and applied controls that were less effective than others. Large scale, computationally intensive, air quality models served a central role in identifying previous misdirections, and continue to provide guidance on what future actions we should take.

Air quality models describe the complex interaction of pollutants between themselves and the environment. For example, ozone, a strong oxidant and primary component of smog, is not emitted directly. Instead, in the presence of sunlight it is formed by chemical interactions between organic gases and nitrogen oxides emitted by a variety of sources, including cars, power plants, manufacturing facilities, paints, etc. At the same time, these pollutants are being transported by the wind, diffusing and depositing to vegetation. This leads to a very complex system of equations, and a typical multidimensional model may require the solution of literally millions of non-linear differential equations. This is, obviously, a computational challenge, one which high performance computing has met.

One of the Grand Challenge Grants in High Performance Computing and Communication, "Distributed Computing in Large Scale Environmental Modeling," a joint team between Carnegie Mellon University and MIT, is developing more physically detailed and computationally efficient air quality models in which to attack the problem. These models, along with high performance computational environments, act as a computational laboratory, allowing scientists to conduct experiments on the computer that would be impossible in the field. For example, it is very difficult, and expensive, to follow pollutant dynamics in the atmosphere above where it is convenient to have surface based measurements. This part of the air is significantly different than near the surface where most of the pollutants are concentrated, but is very important to understanding the formation and fate of many species. For example, such models have been used to look at the formation of acids at night, and the dynamics of radical species whose concentrations are so small it is difficult, if not impossible, to measure them in the air.

One of the central roles of the air quality models is to guide future control directions to improve air quality. For example, the models are being used to investigate the role of alternative fuel use, the types of controls necessary to meet the air quality standards, and how the emissions of one source compare to another in terms of the amount of ozone formed. Regulators are basing their decisions, and billions of dollars in controls, on the outcome of these models. It is important to make sure that they are accurate and comprehensive. Further, it is important to insure that the computational environment is available such that they are exercised fully to identify the best strategies, and not limited by their computational intensiveness. In the Grand Challenge project, both these needs are being met.