EPA/600/A-96/091 Ozone Air Quality Model Development in the U.S. Kenneth Schere1 Atmospheric Sciences Modeling Division Air Resources Laboratory National Oceanic and Atmospheric Administration Research Triangle Park, NC 27711 Air quality models for ozone have been under development and in operation for over twenty years in the United States. We are presently in the third generation of such models. The first generation represented urban scale ozone models, with domain sizes of 100-300 km. The second generation models were regional scale, with domain sizes of 1000-2000 km. The defining feature of third generation models is that they are fully multi-scale, able to simulate the relevant processes leading to ozone accumulation on scales ranging from point source to continental. The key feature of multi-scale models is their ability to nest subdomains of finer grid resolution inside domains of coarser resolution, providing an appropriate level of detail for different portions of the modeling domain. Examples of the various scales within a multi-scale model are: Continental scale Super-Regional scale Regional scale Urban scale Point source scale entire U.S., adjacent regions eastern half of U.S. northeastern U.S. e.g., New York 100 km resolution 40 km resolution 12 km resolution 4 km resolution -0.5 km resolution, or plume-in-grid treatment Typically, the various grid nests are fixed in space for a given application. Adaptive grids, or those that move over time and space to dynamically change resolution as the conditions in the grid change, are being investigated by some groups for application to air quality issues. They are not yet in general use. Figure 1 shows an example of a nested grid configuration for an application of the Urban Airshed Model, Version V (UAM-V, Douglas et al., 1995) to the eastern U.S. The outer grid A contains 40 km grid cells, grid B contains 20 km grid cells, grids C and D - 10 km grid cells, and the urban domains 01, 02, 03 -- 5 km cells. Distinguishing attributes of third-generation models 1. One-way and/or two-way nesting 1 On assignment to the National Exposure Research Laboratory, U.S. EPA, Research Triangle Park, NC 1 ------- Communication between grids of different resolution is a required attribute for multi- scale modeling. Various filtering techniques are in use to prevent spurious waves and mass artifacts to occur near the boundaries of the inner subdomains. Nesting may be one-way, in which the information generated on the coarser grid is used as boundary conditions on the inner (finer resolution) grids. Nesting may also be two-way, where the inner fine grid information is then fed back to the outer coarse grid. Methods for the latter information flow are not always straightforward, as there are no clear-cut routes to aggregate grid information and preserve the correct chemical signal in a non- linear system, such as ozone photochemistry. 2. Greater vertical resolution Another aspect of third generation models is the emphasis on more detailed vertical resolution, especially within the planetary boundary layer. Earlier models, such as the Urban Airshed Model, Versions 4 and earlier (UAM) and the Regional Oxidant Model (ROM), typically contained 3-6 layers in the vertical, extending from the surface to 2-5 km above the surface. Some of the third generation models extend through the full troposphere and contain as many as 30 vertical layers. There is also renewed emphasis on obtaining better near-surface concentration estimates, especially for the nighttime and early morning hours when atmospheric stability causes ozone precursors to accumulate. Layers as shallow as 10 m are now being implemented in some of the models. There are numerical obstacles to coupling a shallow surface layer to a column of cells in a grid model, so off-line or diagnostic techniques are being used to estimate concentrations in the surface layer. 3. Multiple pollutant regimes Previously, separate modeling systems have been developed to address the photochemical ozone regime, fine particle formation, and acid deposition. Within U.S. EPA these (regional scale) models are the ROM, the Regional Particulate Model (RPM), and the Regional Acid Deposition Model (RADM). Third generation models are attempting to incorporate all of the relevant chemical and physical processes to simultaneously handle: • ozone and photochemical oxidants • secondary particle formation (sulfates, nitrates), including size distributions and chemical composition • acid deposition, including aqueous chemistry and full treatment of in-cloud and precipitation systems • toxics (mercury, specific organics) Aside from the obvious savings in computational efficiencies, having these pollutant regimes connected in one modeling system is technically more appealing since most are chemically inter-dependent. Links are also established to feed downstream 2 ------- models, such as water quality models that use atmospheric input as a source term. 4, More chemical detail Increased resolution in chemical detail is also an attribute of third generation air quality models. While lumped structure mechanisms are still in use (such as Carbon Bond 4; -35 chemical species), lumped molecular mechanisms allowing more organic detail are increasingly being used (such as the RADM2 and SAPRC-90, 93) in air quality models. The added detail in the peroxy radical chemistry may make significant improvements in the simulation of oxidant photochemistry. At the same time there is also interest in producing "reduced-form" models, chiefly by simplifying the chemical representations. The motivation for such models is to perform hundreds to thousands of model simulations exploring many combinations of emission control strategies. Since the solution of the gas phase chemistry is typically the largest consumer of computational time in ozone air quality models, a simplified chemical reaction set would be highly desirable. Some attempts have been made to incorporate Graham Johnson's Integrated Empirical Reaction Set (IERS; Johnson and Azzi, 1992), a very compact chemical scheme, into air quality models for ozone. It is yet to be shown whether such models correctly mimic the behavior of models with more comprehensive chemistries over the full range of atmospheric conditions. There have been a number of significant improvements in the numerical solution techniques used for chemical reaction mechanisms as well. "Smarter" schemes are in use that monitor the stiffness of the system in time and space, and change the numerical method to accommodate the demands of the local chemistry. Also, a breakthrough was made in the adaptation of the Gear solution technique (long considered as equivalent to an "exact" solution) to grid models for ozone, by exploiting vector capabilities on some computers. Heretofore the Gear algorithm had been considered as too computationally demanding for use in grid models. 5. Modeling over longer time periods New national ambient air quality standards for ozone and fine particles are expected to be established in the U.S. as early as 1997. It is expected that the new standard for ozone will require modeling analyses over more episodes and for longer averaging periods than the current standard, and the standard for fine particles will require computation of annual averages. The demands on modeling systems will soon become very great for air quality assessments against these new standards. The most direct approach is to explicitly model through the longer time periods required for the new standards. This approach may be prohibitively costly depending upon the number of emissions control modeling scenarios that are needed. Other approaches, based on statistical extrapolation of sets of model results from episodes to a season or a year, are being explored. These approaches rest on the mapping of the 3 ------- representativeness of the episodes to the climatology of the longer time period. 6. Diagnostic analyses The emerging modeling systems also contain various types of new diagnostic analysis techniques that add insights to the basic results that emerge from the model. A few of these techniques are described here. Process Analysis This technique simply tracks the individual contributions of the principal processes to the local change in concentration. One can then determine what the dominant forces are in shaping the concentration patterns at any given place and time in the modeling domain. dc. dc.. < _ y. V dt j dt In the equation above the chemical species are represented as i, and the various processes are represented as j, i.e. j(1)=horizontal transport and diffusion, j(2)=vertical transport, j(4)=vertical diffusion, j(5)=chemical production, j(6)=chemical loss, j(7)=deposition, j(8)=emissions, etc. Each of the partial derivatives is captured and saved during a simulation. Integrated Reaction Rate/Mass Balance The Integrated Reaction Rate/Mass Balance (IRR/MB) technique is a specialized process analysis technique focusing on the chemistry within the model. It tracks initiation, propagation, and termination for radicals and for NO, allowing for better understanding of the NOx and VOC influences on ozone chemistry. Source Attribution Attributing specific portions of the ozone at a given receptor point within a grid model to specific sources or source areas is not a simple problem. Non-linear interactions among the ozone precursors confound the problem of source attribution. Nonetheless, since this is a topic of such intense interest, some empirical techniques are being tested to estimate source attribution. One direct technique is to iteratively modulate the source strength from one source or a group of sources to assess its impact on ozone concentrations. A new technique now being tested in the UAM-X model involves tagging specific VOC 4 ------- and N0X sources and using these tracers along with the ratio of H^g/HNOg production rates as an indicator of NOx-limited or VOC-Iimited chemistry, to apportion source contributions to ozone concentrations. Uncertainty/Error Analysis Methods for assessing sensitivities within models and for propagation of errors through models are being investigated. The ADIFOR technique (automatic differentiation for Fortran) is a generalized technique to track the sensitivities of dependent variables to changes in independent variables in Fortran codes. It is being used to explore sensitivities within models in a global fashion. 7. Higher-level "preprocessors" Meteorology Diagnostic analyses of meteorological observations are rarely used any longer, by themselves, as processors of information for air quality models. Higher-level meteorological forecast models have now been adapted for this use. These are complex models by themselves, requiring at least as many resources for their operation as the air quality models. The most commonly used forecast models in the U.S. for air quality inputs are the NCAR/Penn State MM5 model and the Colorado State RAMS model. These models can resolve complex flows in mountainous terrain or in land/sea situations. Errors which accumulate in such models while operating in a forecast mode can be controlled through four dimensional data assimilation techniques when operating on a historical period as is typical for air quality analyses. Current meteorological models can operate without making hydrostatic assumptions, allowing for more realistic simulations in mountainous terrain or near strongly convective weather events. There are many issues that are still being struggled with in adapting this type of meteorological data to air quality models. Ideally, the horizontal and vertical grid structures of the meteorological and air quality models are identical, allowing direct input of data. Often this is not the case, and the meteorological data must be interpolated to the air quality grid mesh. The possibility of disturbing the dynamic consistency of the met fields exists in this interpolation. Boundary layer parameterizations are not necessarily the same in the models, and sometimes parameters, such as mixing height, must be rederived for use in the air quality model, leading to inconsistencies. Perhaps the most difficult aspect in the linkage is that of clouds, especially the scattered non-precipitating type. In the air quality model these clouds strongly influence photolysis rates, venting of the boundary layer, and chemical processing. Accurate simulation of these clouds is difficult for the meteorological models as there are strong stochastic influences which are not accounted for in deterministic models. Also, 5 ------- the weakly-forced atmospheric system is of prime interest to air quality modelers, and this is exactly the type of system that the meteorological models are least adept at. Coupled models of meteorology and air quality are not yet in fully operational status in the U.S. For the air quality issues being analyzed today, the feedbacks from the chemical system to the meteorological system are not perceived as strong. This may change in the future, as the desire to modulate the incoming radiation with dynamically changing fine particle concentrations (for example) becomes stronger. Emissions Many technical improvements have been made to the emissions modeling systems in use today. Emissions systems such as EMS-95 use GIS-based techniques to better locate emissions sources and allocate them more precisely to model grid locations. Models for biogenic emissions estimates (BEIS2) now have greatly improved emissions factors for isoprene, the most reactive biogenic organic compound. Current isoprene emission flux estimates for the eastern U.S. are 3-7 times greater than earlier estimates of just five years ago. More changes are anticipated as research in the biogenics area continues at an intense pace. The latest models for mobile source emissions estimates include better characterization of running loss and evaporative emissions for VOCs, and updated basic emissions rates and speed correction factors. These latter improvements have increased mobile source emission estimates by 20-30% over earlier model estimates. Problem areas still remain in characterizing the true effectiveness of inspection and maintenance programs and in estimating the impact of "super-emitters" within the automobile population. The next generation of mobile source emissions model will be based on driving modes, and will better characterize different driving patterns on various roadway types (acceleration/power enrichments, cruise, idle, etc.). There will be a need for more city-specific data on roadway links with this type of modal model. Examples of third generation models UAM-V, UAM-X The third generation in the Urban Airshed series of models includes UAM-V (SAI, 1996) and UAM-X (Yocke et al., 1996). The names of these models are actually not correctly descriptive since they are no longer simply urban-scale models. They are multi-scale versions capable of handling air quality simulations from point source scales through super-regional scales. They include Carbon Bond IV chemistry, and are typically driven by versions of the RAMS meteorological model and EMS-95 emissions model. They include plume-in-grid treatments for major point sources, 6 ------- based on a version of the Reactive Plume Model. These models are principally used for ozone episodes characterized by attendant high pressure, clear to partly cloudy skies, and lack of widespread precipitation. The models lack full treatment of clouds, dynamics, and aqueous processes which poses an obstacle for more general applicability beyond photochemical oxidants. The UAM-V model is fully operational and has been used in several major U.S. modeling applications, including the OTAG (Ozone Transport Assessment Group) eastern U.S. simulations for episodes during 1988, 1991, and 1995. URM The Urban to Regional Multi-scale Model (URM; Kumar and Russell, 1996; Kumar et al., 1994) is a multi-scale version of the urban photochemical grid model originally developed at the California Institute of Technology by John Seinfeld and colleagues, known as the CIT model. The URM is similar in structure to the UAM-V in that its vertical dimension extends through and slightly above the boundary layer. It can model various configurations of nested grids, in one- and two-way nested modes, using a finite-element formulation that does not require regularly spaced mesh points. The model uses the LCC chemical mechanism, a predecessor of the SAPRC mechanism and includes only gas-phase chemistry. The URM can be driven by various types of meteorological and emissions models. The inputs must be interpolated to the URM grid structure. A reactive plume model has recently been developed, using various levels of chemical detail, for use within the URM model (Kumar and Russell, 1996). Major applications have been for the Los Angeles/SCAQS database and the northeastern U.S. SAQM The SARMAP Air Quality Model (SAQM; Chang, 1996) is a direct descendent of the Regional Acid Deposition Model (RADM). While the RADM is principally a single- scale model having been used at super-regional and regional scales, the SAQM is a nested multi-scale model capable of both one- and two-way nesting. (Note that a research version of RADM using 1-way nesting has been in existence since 1989.) The depth of the modeling domain extends through the troposphere into the lower stratosphere. The formulation contains treatments for clouds, aqueous chemistry, and precipitating systems. (Initial applications of SAQM, however, have been run "dry", i.e., no clouds or aqueous processes, for the California San Joaquin study, SARMAP. SAQM is now being configured for eastern U.S. applications allowing for wet processes.) There are versions of the SAQM for various chemical mechanisms, including RADM2, SAPRC-93, and Carbon Bond-4. The SAQM is driven by the MM5 meteorological model and uses the same horizontal and vertical coordinate systems as the MM5. A surface layer on the order of 10m depth is implemented in both SAQM and MM5. The SAQM applications have been driven by the EMS-95 emissions model. Point source treatment is handled by a series of telescoping grids surrounding 7 ------- the source. The model is capable of addressing the acid deposition and photochemical oxidants issues, and is currently being extended to include size- distributed fine particles as well. Models-3, MAQSIP Another descendent of the RADM model is the EPA's Models-3, or third generation nested air quality model (Dennis et al., 1995; Byun et al., 1995). EPA's research collaborator in the model development effort, MCNC, calls their version of the model Multiscale Air Quality Simulation Platform (MAQSIP). The Models-3 is in final development and testing stages and should be operational toward the end of 1997. The model is multi-scale with nested grids in a generalized coordinate system. Many of the technical features are similar to the SAQM model, including the ability to use non-hydrostatic meteorological information coming from MM5 in the air quality calculations, detailed cloud and aqueous chemistry processes, and multi-pollutant capabilities including acid deposition, photochemical oxidants, and fine particles. Distinguishing features of Models-3 include a high-level user interface, facilitating user access to the complex modeling system, a greater degree of modularity at the simulation process level than has existed in earlier models, and linked analysis and visualization subsystems. An example of the modularity is the generalized chemical mechanism reader and general numerical solver, capable of handling any gas-phase chemical kinetic mechanism. Chemical mechanisms tested thus far include the RADM2 and the Carbon Bond-4. Future plans include the SAPRC mechanism and ultimately a version of the emerging "morphecule" chemical mechanism. Driver systems include the MM5 meteorological model and the EMS-95 emissions model. Initial evaluations will be accomplished using eastern U.S. data sets, including the Southern Oxidants Study- Nashville field campaign data. References Byun, D., A. Hanna, C. Coats, D. Hwang, 1995: Models-3 air quality model prototype science and computational concept development, Transactions, AWMA Specialty Conference on Regional Photochemical Measurement and Modeling Studies, Nov. 8- 12, 1993, 197-212. Chang, J., 1996: User's Guide to the SARMAP Air Quality Model (SAQM), California Air Resources Board, Sacramento, CA, in press. Dennis, R., D. Byun, J. Novak, K. Galluppi, C. Coats, M. Vouk, 1995: The next generation of integrated air quality modeling: EPA's Models-3, Atmos. Environ., Vol. 30, 1925-1938. Douglas, S., G. Mansell, T. Myers, M. Jimenez, D. Perry, 1995: Application of UAM-V 8 ------- to the Northeast as Part of Phase I of the Modeling Ozone Cooperative (MOCA), Report SYSAPP-95/049, Systems Applications Intl., San Rafael, CA. Johnson, G.M., M. Azzi, 1992: Notes on the derivation: the integrated empirical rate model (V2.2), CSIRO Report, Division of Coal and Energy Technology, North Ryde, Australia. Kumar, N., A. Russell, 1996: Multiscale air quality modeling of the northeastern United States, Atmos. Environ., Vol 30, 1099-1116. Kumar, N., A. Russell, 1996: Development of a computationally efficient, reactive subgrid-scale plume model and the impact in the northeastern United States using increasing levels of chemical detail, Jour. Geophys. Res., Vol. 101, 16737-16744. Kumar, N., M. Odman, A. Russell, 1994: Multiscale air quality modeling: Application to southern California, Jour. Geophys. Res., Vol. 99, 5385-5397-16744. SAI, 1996: User's Guide to the Variable-Grid Urban Airshed Model (UAM-V), Report SYSAPP-95/027, Systems Applications Intl., San Rafael , CA. Yocke, M., R. Morris, G. Yarwood, S. Shepard, T. Steockenius, 1996: The Extended Urban/Regional Airshed Model (UAM-X), in Proceedings of 89th AWMA Annual Meeting, Nashville, TN, June 23-28, 1996. Disclaimer This paper has been reviewed in accordance with the U.S. Environmental Protection Agency's peer and administrative review policies and approved for presentation and publication. Mention of trade names or commercial products does not constitute endorsement or recommendation for use. 9 ------- Deg. Longitude -89 -94 -84 -79 -74 -69 47 40 43 30 39 « 20 10 31 50 10 20 40 30 Figure 1. An example of a nested grid configuration for an application of the UAM-V model to the eastern U.S. 10 ------- TECHNICAL REPORT DATA ¦ X. REPORT SO. EPA/600/A-96/091 2. 3.RECIPI! 4. TITLE AND SUBTITLE Ozone Mr Quality Model Development in the U.S. S.REPORT DATE 6.PERFORMING ORGANIZATION CODE 7. AtJTHOR(S) Kenneth Schere 8.PERFORMING ORGANIZATION REPORT NO. 9. PERFORMING ORGANIZATION NAME AND ADDRESS Same as Block 12 10.PROGRAM ELEMENT NO. 11. CONTRACT/GRANT NO. 12. SPONSORING AGENCY NAME AND ADDRESS National Exposure Research Laboratory- Office of Research and Development U.S. Environmental Protection Agency Research Triangle Park, NC 27711 13.TYPE OF REPORT AND PERIOD COVERED Proceedings, FY-97 14. SPONSORING AGENCY CODE EPA/600/9 15. SUPPLEMENTARY NOTES 16. ABSTRACT Air quality models for ozone have been under development and in operation for over twenty years in the United States. We are presently in the third generation of such models. The first generation represented urban scale ozone models, with domain sizes of 100-300 km. The second generation models were regional scale, with domain sizes of 1000-2000 km. The defining feature of third generation models is that they are fully multi-scale, able to simulate the relevant processes leading to ozone accumulation on scales ranging from point source size to continental. The key feature of multi-scale models is the ability to nest subdomains of finer grid resolution inside domains of coarser resolution, providing an appropriate level of detail for different portions of the modeling domain. 17. KEY WORDS AND DOCUMENT ANALYSIS a. DESCRIPTORS b. IDENTIFIERS/ OPEN ENDED TERMS c.COSATI 18. DISTRIBUTION STATEMENT RELEASE TO PUBLIC 19. SECURITY CLASS (This Report) UNCLASSIFIED 21.NO. OF PAGES 20. SECURITY CLASS (This Page) UNCLASSIFIED 22. PRICE ------- |