FEATURE
Air Quality Modeling's Brave New World A new generation of software systems is set to tackle regional and multipollutant air quality issues. ELAINE
L.
W
hen Charlotte, North Carolina, was designated a nonattainment area for o z o n e u n d e r t h e Clean Air Act Amendments of 1990, Brock Nicholson staged a full frontal attack on the problem. Nicholson, assistant chief for planning in the Air Quality Section of the North Carolina Department of Environment, Health and Natural Resources did what remains only a dream for most air quality regulaand scientists Starting in 1993 he performed h u n d r e d s of model runt: on a f^ray "try out" different proposed emissions control strategies His " w h a t if" modeling runs p l u t w d control s t r a t e g i e s snrh as reHiirinu noint that simulate ozone formation under various weather, geographical, and chemical conditions. He used the standard EPA-sanctioned urban ozone model, Urban Airshed Model 4 (UAM-4), and an early prototype of the Environmental Decision Support Systern (EDSS)r a modeling system that will run on everything from supercompters to work stations. The modeling results led Nicholson and his colleagues to NO x control strategies that seemed scientifically sound and cost effective. EPA approved the state implementation plan containing those strategies last year. Nicholson and his colleagues' work was done at the North Carolina Supercomputer Center (NCSC) in Research Triangle Park, where researchers funded by EPA were developing EDSS. 2 0 0 A • VOL. 30, NO. 5, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY / NEWS
APPLETON
So massive is the computing power needed to run photochemical grid models that local regulators without supercomputer resources are lucky if they can run two or three air quality simulations. Today, modeling is more often used to demonstrate attainment of ozone regulations than to develop potential control strategies. "The use of models to design control strategies has historically been underutilized. This has been primarily due to the computing cost to run p h o t o c h e m i c a l grid models," wrote Neil J. M. Wheeler, manager of the Control Strategy Modeling Section, California Air Resources Board (i). But Nicholson's "what-if" approach will be the wave of the future if model developers in EPA labs have their way. Since 1992, EPA has been creating a new generation of software—Models-3—that is widely regarded as the next-generation air quality modeling system. The system has a modular framework that allows users to integrate a broad variety of air quality models. In the future, users will also be able to plug in economic decision support tools. A prototype version of Models-3 already exists in the Atmospheric Modeling Division of EPA's National Exposure Research Laboratory in Research Triangle Park EDSS was developed as a rapid prototype of Models-3 under a three-vear $7 8 million cooperative agreement with EPA An operational version of Models-3 may be in the hands of scientists and state air quality regulators by late 1997. But a preliminary or beta version may be distributed early in 1997 so users can provide feed0013-936X/96/0930-200A$12.00/0 © 1996 American Chemical Society
Modeling a control strategy with Models-3 A sample of EPA's Models-3 display screen shows how a state environmental policy analyst can study the effects of an emissions control strategy on several pollution problems simultaneously. The blue area depicts tropospheric ozone levels that meet existing National Ambient Air Quality Standards but exceed one of the possible revised standards. The green areas show high sulfur dioxide concentrations directly over large point sources. Yellow plumes depict sulfate, which was transformed from S02 and transported by winds. Because the chemistry of ozone and sulfate formation in the atmosphere are linked changes in source emissions influence and sulfate concentrations Source: EPA Scientific Visualization Center Research Triangle Park N C
back to developers. Like EDSS, the system runs on a variety of computer platforms, including a Cray T3D, one of the world's most powerful supercomputers. Models-3 promises to eventually furnish regulators with more reliable air quality management information than current models in far less time and with greater ease. In particular, the system will incorporate models that use regional-to-urban scales. These will include nested grids that allow modelers to use a fine scale to simulate urban air quality within larger cells used for regional simulations. With this technology, regulators can interactively try out regional control strategies alongside the urban-only simulations that are in widespread use today. Models-3's modular design lets users can plug in different chemical, weather, and geographical models according to their needs. They can replace old models with state-of-the-art versions as new algorithms and models become available. Ultimately, the system will incorporate visualization software so that users can work interactively with three-dimensional displays of model output (2). Developers hope the new, more user-friendly system will make it easier to run models and present information to policy makers in graphical ways that are easy to understand. "We've been feeding information to decision makers for years," said California's Wheeler. He has been watching the Models-3 de-
velopment closely and contributing experience from the California Air Resources Board's 20 years of ozone modeling. "We've become aware that a scientist giving a few numbers and some cryptic plots is not always the way to communicate information to a decision maker." In addition, Models-3 will ultimately become a socalled "comprehensive modeling system" that enables users to simulate pollutants in other media, such as water. EPA also plans to include models that simulate health effects and other pollution consequences, said loan Novak, the Models-3 project manager and chief of the modeling systems branch of the Atmospheric Modeling Division at the National Exposure Research Lab.
No magic bullet Some air quality managers have reservations about the idea of a grand, comprehensive environmental management system. They think that even today's more limited models are used incorrectly, noting that some regulators rely on them without enough consideration of their inherent uncertainty. In California, Wheeler has issued guidelines on uncertainty to regulators using air quality models. "The ones who want to know the uncertainty are the real planners," he said. "The others who don't want to know are the people just meeting regulatory requirements—[those who say] we have to do the modeling because EPA requires it." VOL. 30, NO. 5, 1996/ENVIRONMENTAL SCIENCE & TECHNOLOGY / NEWS • 2 0 1 A
Before EPA sanctions the use of Models-3, the system will undergo six to nine months of scientific verification, said Novak. Even so, EPA is "busily drawing up guidelines for regulators" on how to use models qualitatively, not quantitatively, said Bachmann. "In our lifetime, models are going to be approximations of the truth but not at all the truth," he said. That is not to say Bachmann is not in favor of developing better tools such as Models-3. "I'm sending a strong note of support," he added, "but the new religion of the regulatory side is to make sure we don't overuse models."
Tackling the regional dilemma
Visualizing ozone control strategy alternatives The effects of alternative ozone control strategies can be graphically illustrated and communicated to policy makers with images such as this one produced in January using the Environmental Decision Support System. Produced after a twohour supercomputer run, the image shows ground-level ozone concentrations over central California; it was based on atmospheric chemical and meteorological data for Aug. 3,1990, 4 p.m. local time. The model can demonstrate to policy makers, for instance, the overall effects of reducing NO, emissions from big point sources or motor vehicles in urban areas. The red areas of high ozone (0.100 parts per million by volume) are rural regions that have no big point sources but experience the effects of atmospheric transport. The regions were upwind of urban areas with high N0 emissions. Source: North Carolina Supercomputing Center.
The Clean Air Act mandates that certain ozone nonattainment areas use an EPA-sanctioned photochemical grid model (the de facto standard is Urban Airshed Model-4, UAM-4) to demonstrate that proposed emissions control strategies will enable attainment. Because a multitude of complex factors influence an emissions reduction strategy, some regulators, naturally, turn to computers for the right answers. EPA and other regulators have tended to rely on models as "magic bullets" rather than using them as tools. "We as regulators have asked too much in the past and expected too much of models," said John Bachmann, associate director for science policy at the EPA Office of Air Quality Planning and Standards in Research Triangle Park. "Unfortunately I think a lot of people in the modeling community felt obliged to deliver what can't be delivered, which is a level of certainty that is just not possible." 2 0 2 A • VOL. 30, NO. 5, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY / NEWS
The new system is intended to overcome acknowledged flaws in today's models and help regulators keep up with changes in science and environmental policy. For instance, a single state rarely produces all the tropospheric ozone found above its boundaries. Ozone in the lower atmosphere is a regional problem, but until recently, EPA considered states to be responsible for its reduction. Not so today. EPA and state regulators are recognizing the need for regional and possibly even nationwide ozone attainment strategies. In 1994, the 32state Ozone Transport Assessment Group (OTAG) was formed to explore regional and interregional ozone transport in the eastern United States after states were unable to demonstrate attainment with local controls. OTAG will propose to EPA new regional attainment levels, which will ultimately be imposed into local models. "We might be able to drop the cost nationwide of meeting the ozone standard if we make rational reductions in regional ozone levels, and through that propose less onerous local strategies," said EPA's Bachmann The lack of comprehensive regional modeling tools is a major obstacle to developing regional strategies. Three primary ozone models are in use today: the UAM-4, the Regional Oxidant Model (ROM), and the Urban Airshed Model-V (UAM-V), a variablescale model. But only UAM-V uses a regional-tourban grid that can model pollutants on the wide range of scales necessary to examine ozone locally and regionally. And UAM-V is not yet in wide use. One reason California does not use it, said Wheeler, is because it is proprietary and thus cannot be delivered to consultants working for the state. UAM-4, with no regional component, cannot help modelers study regional-to-urban issues. "We now recognize that ozone is not a local problem, it's a regionwide problem. And that's where UAM-4 falls down," said S. T. Rao, technical chair of the Ozone Transport Commission's modeling committee and director of the Bureau of Air Research, New York State Department of Environmental Conservation. More typical is the use of the ROM for regional modeling, but some regulators say its boundary simulations are questionable. The Massachusetts Dep a r t m e n t of Environmental Protection, for instance, uses the ROM to simulate transport of ozone into New England from heavily populated regions to the southwest. One problem, said SteVe Dennis, chief of the Division of Air Quality Control Modeling Branch, is that "we just don't have accurate future-
year boundary conditions. We don't yet know what people are going to pick for future strategies." For instance, in New England, the ROM significantly overpredicts ozone concentrations when compared to field data from two episodes in July 1988, he said. Massachusetts uses these episodes among others for the New England attainment demonstration. It is a technological problem with significant political ramifications. "Boundary conditions are particularly important to simulate, because we're really pointing to our neighbors to the south of us as causing much of our problem, and our neighbors to the north are doing the same thing," Dennis said. Additionally, most models available to regulators simulate only one pollutant, in this case ozone. "We don't have any modeling system that would allow us to look at regional scale transport of fine particulate matter and that would predict regional haze and visibility problems," said New York's Rao. "Today, people are not dealing with [regional haze and fine particulate matter] because they don't know what to do." EPA is obligated, he said, to issue a fine particulate standard and to propose an alternative ozone standard (3).
A flexible design for credible modeling Designers view Models-3 as a framework for continuous improvement of modeling science that will permit wide use of more credible models. A modular design gives Models-3 great flexibility to adapt to changes in science, technology, and the needs of individual users. Users can create a customized executable model by selecting any of a variety of available science modules. They also can incorporate into the system a completely independent model, such as a meteorological model. Users have a choice of chemistry mechanisms, including a general chemistry reader and several well-known atmospheric chemistry mechanisms. They can select any of several advection and diffusion schemes that control the horizontal and vertical motion of the atmosphere. Models-3 can use several types of public domain, commercial, and custom visualization-analysis packages.
More monitoring data needed Moreover, there are widely acknowledged defects in the emissions databases used as input to the models. As little as two years ago, people were worried that there was a factor of two emissions error," said EPA's Bachmann. For instance, regulators struggle to simulate emissions from organic matter. "Right now modelers are using vegetation emissions generated from a model called BEIS 1," said Massachusetts' Dennis. "It appears to be underestimating emissions from trees." More egregiously, until recently modelers relied on a 20-year-old plume database for information about large point sources, said Bob Imhoff, a systems analyst and atmospheric modeler with the Tennessee Valley Authority in Muscle Shoals, Ala. That was a serious flaw: large point sources, like TVA's coal-fired power plants, are responsible for more than half of the NO emissions in the country, he stated. Not only do holes in the emissions databases cause problems with input, but lack of real data hinders model validation. A 1991 National Academy of Sciences report on tropospheric ozone (4) concluded that "The air quality management system was theoretically sound but flawed in practice because we hadn't done the feedback," said Bachmann. In other words, said the NAS, researchers were not doing enough field monitoring after control strategies were implemented to determine whether the models had actually predicted true outcomes. Research efforts, including those of EPA, OTAG, and the North American Research Strategy on Tropospheric Ozone (NARSTO), are today collecting additional air quality field data needed to improve future models. "We're trying to expand our envelope to use real air quality data wherever possible," said Bachmann. If regulators are unhappy with current models, industry is even more disturbed. As control technology costs mount, utilities and other major polluters say they
want to ensure that the strategies they are mandated to implement are based on sound science.
The promise of Models-3 Built using a client-server structure that allows users to tap into powerful computers from a pointand-click graphical front end, Models-3 is designed to adapt to the evolution not simply of air quality modeling systems but of other environmental models and economic decision support tools as well. To plan for model improvements and the advent of cross-media modeling, the EPA lab is using an object-oriented approach. It will allow users to "unplug" obsolete or ineffective models and plug in new ones, said Models-3 Project Manager Novak. "You don't have to throw away the whole system," she added. "We can continue to upgrade components within the existing framework." Today the system's graphical user interface, intended to make it easier for the user to perform simulations, runs on Sun, Silicon Graphics, Digital Equipment Corp. Alphas, and other UNIX workstations. Servers range from clusters of powerful workstations on a network to a Cray C90 supercomputer at EPA's National Environmental Supercomputing Center in Bay City, Mich. "It can be configured in very different ways," said Novak, "but the idea is that if you've got a large enough problem, you want different servers for different pieces of the problem." Developers are also tuning the parts of the system currently the meteorological models to run in parallel on a 128-processor Cray T3D supercomputer cicquired lcist summer by the Bay City center Part of the impetus for designing a hardwareneutral system was to allow EPA to distribute code to users to run on whatever systems they have available, said Novak. "We plan to give Models-3 away to anyone who wants it." VOL. 30, NO. 5, 1996/ENVIRONMENTAL SCIENCE & TECHNOLOGY / NEWS " 2 0 3 A
Proposed ozone standard challenges technology A computer system with the capabilities of Models-3 may be needed if EPA adopts a proposed new ozone standard. The current standard, 0.12 ppm measured in one-hour episodes, is being revisited by EPA after a 1993 lawsuit by the American Lung Association. The ALA asserts that the current standard ignores ozone's harmful cumulative effects on the lungs of children and outdoor workers. EPA is now considering resetting the standard to as low as 0.070.09 ppm and measuring in eight-hour episodes. Storing and manipulating eight hours' worth of air quality data require an enormous amount of computer memory, disk space, and speed. If EPA adopts an eight-hour standard, regulators would need a more powerful modeling system to record and simulate longer episodes. Models-3 could be the framework for that simulation, predicts John Bachmann of EPA's Office of Air Quality Planning and Standards. The Environmental Decision Support System, an early Models-3 prototype, is already proving this kind of computing to be possible. The Southeast States Air Resources Managers will use the system, running on a Cray supercomputer, to simulate the entire mid-May to mid-September ozone season, using 24-hour days. "We'll create an ozone climatology for the Southeast," said North Carolina Supercomputer Center Director of Advanced Applications Ken Galluppi. "That's never been done before because we've never had the computer power, and we've never had a system that could handle this much information." —ELAINE APPLETON
Despite the promise of "plug-and-play" hardware and software modules, Novak acknowledged that initial implementations of Models-3 will only loosely integrate model input and results. "There's no one tool that can do everything," she said. "We're trying to provide a framework that can help people access a variety of data and pull different data sets together for analysis." She hopes users will be able to run different models on such integrated data sets. In theory, that would provide better comparisons between model output than is possible with today's disparate and incompatible data sets. Practical use of Models-3 as a tool for designing and demonstrating regional control strategies depends on availability of a master database of air quality model results. EPA is working with regional and local organizations to build the database and make it available to the environmental community. NARSTO, for example, plans to contribute its own field data to the Models-3 database, said Al Farewell, a NARSTO committee chair and atmospheric scientist with Pennsylvania Power and Light in Allentown, Pa. Novak suggested that states doing their own independent modeling could share the results with, say, the Ozone Transport Commission by sending model results to the EPA's master database. The idea is to have an automated way to catalogue and make available the huge amounts of information generated by these models. Models-3 is not the only comprehensive modeling effort under way. Canada is developing such a system. So too is the Electric Power Research Institute (EPRI) with a similar effort it calls CAMRAQ (5). EPRI, based in Palo Alto, Calif., is largely funded by utilities and other regulated industries. EDSS will also be available, in some fashion, 2 0 4 A • VOL. 30, NO. 5, 1996 / ENVIRONMENTAL SCIENCE & TECHNOLOGY / NEWS
through the NCSC—probably before Models-3 is fully operational. The system currently provides three capabilities: model engineering and configuration management services; simulation planning and execution management; and analysis and threedimensional visualization (6). In the future, EDSS will include other functions such as surface and groundwater modeling; cross-media simulations of air, water, and land; economic models; and health effects models. Like Models-3, EDSS already provides urban and regional ozone models, and it will support multiscale particulate and acid deposition modeling. In June 1995, NCSC released a beta version of EDSS to pilot testers, including EPA, Cray Research, and university labs; test versions were released to state and local regulators in January. The center is offering an automated version of the Urban Airshed Model, called UAM Guides, on the World Wide Web. EDSS itself is too large to distribute via the Internet, officials said. Although new systems like Models-3 and EDSS are designed to make it easier to perform model runs, experts like Wheeler believe such ease of use could cause as many problems as it solves. "It becomes easier for someone to run the model and not necessarily be cognizant of the problems in it," he said. One solution, said NCSC officials, is to provide education as an integral part of any modeling program, a response endorsed by Wheeler. "We've learned a lot over years of experience in making mistakes, and part of the education is explaining to people what mistakes we've made and what to look for." A critical part of that educational process, he stressed, is to make users aware of degrees of uncertainty. "We know there will always be uncertainty in emissions, meteorology, model formation, et cetera," said Wheeler. "We need to tell people how much confidence they should have in a model, because the models don't make decisions, the decision makers do."
References (1) Wheeler, N.J.M. "Regional/Urban Air Quality in California, Part I: Air Quality, Simulation, and Decision Making in California"; Proceedings of the Conference on Environmental Impact Prediction: Simulation for Environmental Decision-Making, October 6-7,1994; MCNC, North Carolina Supercomputer Center: Research Triangle Park, NC, 1995. (2) Dennis, R. L. et al. Atmos. Environ. 1996, April. (3) Reichhardt, T. Environ. Sci. Technol. 1996, 30, 68A. (4) Rethinking the Ozone Problem and Urban and Regional Air Pollution; National Academy of Sciences: Washington, DC, 1991. (5) Hansen, D. A. et al. Environ. Sci. Technol. 1994, 28, 71A. (6) Ambrosiano, J. et al. Presented at EPA Next-Generation Environmental Modeling Computational Methods Workshop, Bay City, MI, Aug. 7-9, 1995.
Elaine L. Appleton is a freelance science writer based in Newburypor,, Mass. She is former senior editor at Datamation magazine.
An electronic version of this article is accessible via the World Wide Web on the American Chemical Society's Publications home page (http://pubs.acs. org/hotartcl/index.html). An animation of the Models-3 screen shown on page 201A is included as a Quicklime movie and an MPEG file. —Editor
