Page 303

10
Ozone Air-Quality Models

Introduction

To predict compliance with the ozone air-quality standard at some future date it is necessary to know how ozone concentrations change in response to prescribed changes in source emissions of precursor species: the oxides of nitrogen (NOx) and volatile organic compounds (VOCs). This assessment requires an air-quality model, which in the case of ozone prediction is often called a photochemical air-quality model. The model in effect determines the emission reductions needed to achieve the desired air-quality standard, such as the National Ambient Air Quality Standard (NAAQS) for ozone.

Air-quality models are mathematical descriptions of the atmospheric transport, diffusion, and chemical reactions of pollutants. They operate on sets of input data that characterize the emissions, topography, and meteorology of a region and produce outputs that describe that region's air quality. Mathematical models for photochemical air pollution were first developed in the early 1970s and have been developed, applied, and evaluated since that time. Much of the field's history is described in reviews by Tesche (1983), Seinfeld (1988), and Roth et al. (1989).

The air-quality model is theoretically the ultimate integrator of one's knowledge of the chemistry and physics of the ozone-precursor system. Photochemical air-quality models can be used to demonstrate NAAQS attainment or to educate planners about the emissions controls needed to head toward attainment. Whether or not they are actually used in determining abatement



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 303
Page 303 10 Ozone Air-Quality Models Introduction To predict compliance with the ozone air-quality standard at some future date it is necessary to know how ozone concentrations change in response to prescribed changes in source emissions of precursor species: the oxides of nitrogen (NOx) and volatile organic compounds (VOCs). This assessment requires an air-quality model, which in the case of ozone prediction is often called a photochemical air-quality model. The model in effect determines the emission reductions needed to achieve the desired air-quality standard, such as the National Ambient Air Quality Standard (NAAQS) for ozone. Air-quality models are mathematical descriptions of the atmospheric transport, diffusion, and chemical reactions of pollutants. They operate on sets of input data that characterize the emissions, topography, and meteorology of a region and produce outputs that describe that region's air quality. Mathematical models for photochemical air pollution were first developed in the early 1970s and have been developed, applied, and evaluated since that time. Much of the field's history is described in reviews by Tesche (1983), Seinfeld (1988), and Roth et al. (1989). The air-quality model is theoretically the ultimate integrator of one's knowledge of the chemistry and physics of the ozone-precursor system. Photochemical air-quality models can be used to demonstrate NAAQS attainment or to educate planners about the emissions controls needed to head toward attainment. Whether or not they are actually used in determining abatement

OCR for page 303
Page 304 strategies, models are essential for examining of the complex interactions among emissions, meteorology, and atmospheric chemistry. A practical model consists of four structural levels: • The conceptual formulation; that is, a set of assumptions and approximations that reduce the actual physical problem to an idealized one, which, within the limits of present understanding, retains the most important features of the actual problem. • The basic mathematical relations and auxiliary conditions that describe the idealized physical system. • The computational schemes (numerical procedures) used to solve the basic equations. • The computer program or code that actually performs the calculations. The term ''model'' has been used to apply collectively or separately to all four levels. Models of a particular process, or a group of interacting processes, are called component models or modules. The basis for air-quality models is the atmospheric diffusion equation, which expresses the conservation of mass of each pollutant in a turbulent fluid in which chemical reactions occur (Seinfeld, 1986). For at least a decade, the Environmental Protection Agency (EPA) has offered guidelines on the selection of air-quality modeling techniques for use in State Implementation Plan (SIP) revisions, new source reviews, and studies -aimed at the prevention of significant deterioration of air quality. EPA guidelines (EPA, 1986b) identify two kinds of photochemical model: The urban airshed model (UAM) is the recommended model for modeling ozone over urban areas and EKMA (empirical kinetic modeling approach) is identified as an acceptable approach. As noted in Chapter 3, the 1990 Clean Air Act Amendments specify that three-dimensional, or grid-based, air-quality models, such as UAM, be used in SIPs for ozone nonattainment areas designated as extreme, severe, serious, or multistate moderate (EPA, 1991b). Grid-based models use a fixed Cartesian reference system within which to describe atmospheric dynamics (Seinfeld, 1988). The region to be modeled is bounded on the bottom by the ground, on the top by the inversion base or some other height that characterizes the maximum extent of vertical mixing, and on the sides by east-west and north-south boundaries, unless the coordinates are rotated. This space is then subdivided into a three-dimensional array of grid cells. The horizontal dimensions of each cell are usually a few kilometers for urban applications. Some older grid-based models assumed only a single, well-mixed vertical cell extending from the ground to the inversion base; current models subdivide the re-

OCR for page 303
Page 305 gion into layers. Vertical dimensions can vary, depending on the number of vertical layers and the vertical extent of the region being modeled. A compromise generally must be reached between the better vertical resolution afforded by the use of more vertical layers and the associated increase in computing time. Although aerometric data, such as the vertical temperature profile, that are needed to define the vertical structure of the airshed are generally lacking, it is important to use enough vertical layers so that NOx emissions from tall stacks are not overdiluted computationally. There are practical and theoretical limits to the minimum horizontal grid cell size. Increasing the number of cells increases computing and data acquisition effort and costs. In addition, the choice of the dimension of a grid cell implies that the input data—information about winds, turbulence, and emissions, for example—are resolved to that scale. In practice, most urban models use horizontal grid cell of a few kilometers, whereas regional models use horizontal grid cells of tens of kilometers. There is a need for a set of directives about how to specify the size of the modeling domain, the horizontal grid spacing. to be used, the vertical extent of the modeling region, and the number and resolution of vertical layers. These directives must be based on the exercise of models having a wide range of spatial resolutions and on the comparison of model performance against a wide variety of high-quality field data. It has been found that increasing the horizontal grid spacing in a photochemical air-quality model will generally result in a reduction in the peak ozone concentration. It also is important to provide adequate vertical resolution—the order of five vertical layers or more for urban-scale applications (EPA, 1991b). The minimum amount of meteorological and air-quality data must be prescribed as modeling inputs. The choice of the size of the modeling domain will depend on the resolution available in the data, including the distribution of emissions in the region, the weather conditions, and, to some extent, the computational resources available. The spatial resolution of the concentrations predicted by a grid-based model corresponds to the size of the grid cell. Thus, effects that have spatial scales smaller than those of the grid cell cannot be resolved. Such effects include the depletion of ozone by reaction with nitric oxide (NO) near strong sources of NOx like roadways and power plants. Several grid-based photochemical air-quality models have been developed to simulate ozone production in urban areas or in larger regions. They differ primarily in their treatment of atmospheric processes and in the numerical procedures used to solve the governing system of equations. Table 10-1 lists grid-based photochemical air-quality models in current use or under development. As noted in Chapter 3, photochemical air-quality models are used in determining the emissions controls needed to attain the ozone NAAQS. But

OCR for page 303
Page 306 TABLE 10-1 Photochemical Air Quality Models Model Developer Status Reference UAM (Urban Airshed Model Systems Applications, Inc. Wide use Documented Reynolds et al. (1973, 1989 Carbon Bond-II Chemistry     Reynolds (1977)       Ames et al. (1985) UAM Systems Applications, Inc. Limited   Carbon Bond-IV Chemistry       UAM Radian Corporation In testing Tesche et la. (1988a,b) Carbon Bond-IV Chemistry   Code available Tesche and McNally (1989) CIT (California Institute of Technology) California Institute of Technology and Carnegie Mellon Research applications McRae et al. (1982) CIT Chemistry University Not documented McRae and Seinfeld (1983)       Russell and Cass (1986) CIT Carnegie Mellon Limited use   SAPRC (Statewide Air Pollution Research Center) Chemistry University     (Table continued on next page)

OCR for page 303
Page 307 (Table continued from previous page) Model Developer Status Reference CALGRID Sigma Research In review Yamartino et al. (1989) SAPRC Chemistry   Documented   ROM EPA Operational (code and documentation due in 1990) Lamb (1983) Carbon Bond-IV Chemistry       RADM (Regional Acid Deposition Model) National Center for Atmospheric Research and SUNY Albany Operational Chang et al. (1987) ADOM (Acid Deposition and Oxidant Model) ENSR and Ontario Operational Venkatram et al. (1988)   Ministry and Environment    

OCR for page 303
Page 308 first, the validity of the model must be demonstrated by its ability to simulate adequately a base-year episode of high concentrations of ozone. Then, using the same meteorology as in the base-year episode, the emissions are hypothetically reduced to the point at which the peak 1-hr ozone concentration in the region does not exceed 120 ppb. The remainder of this chapter is structured as follows. First the development of meteorological inputs to air-quality models is discussed, followed by a brief discussion of boundary and initial conditions. We then address how attainment of the ozone NAAQS is demonstrated using a grid-based modeling approach. Because urban grid-based photochemical air-quality models have received a great deal of attention in the literature (see, for example, Seinfeld, 1988), we do not devote significant coverage to them here. However, regional air-quality models, which are quite similar in structure to the earlier models used in urban applications, have not undergone the same degree of evaluation and application as their urban counterparts. For this reason, and because such regional models are currently being used to assess ozone abatement strategies in areas like the northeastern United States, we will review regional grid models in this chapter. The particular model on which we focus is EPA's Regional Oxidant Model (ROM). After the analysis of regional models, the general issue of evaluation of model performance is addressed. Meteorological Input To Air-Quality Models Grid-based air-quality models require, as input, the three-dimensional wind field for the episode being simulated. This input is supplied by a so-called meteorological module. Meteorological modules for constructing wind fields for air-quality models fall into one of four categories (Tesche, 1987; Kessler, 1988): • Objective analysis procedures that interpolate observed surface and aloft wind speed and direction data throughout the modeling domain. • Diagnostic methods in which the mass continuity equation is solved to determine the wind field. • Dynamic, or prognostic, methods based on numerical solution of the governing equations for mass, momentum, energy, and moisture conservation along with the thermodynamic state equations on a three-dimensional, finite-difference mesh. • Hybrid methods that embody elements from both diagnostic and prognostic approaches.

OCR for page 303
Page 309 Objective analysis procedures are inexpensive and simple to use. Their disadvantage is that they contain no physics-based calculations, and the results are highly dependent on the temporal and spatial resolution of the observed wind speeds and directions. Results are often unsatisfactory in areas of the modeling domain where observations are either sparse or not representative of the physical geography. Areas of complex terrain, variations in land use, and ocean-land contrasts cannot be accounted for. Diagnostic procedures impose mass consistency on the flow field through appropriate equations, and can crudely include terrain blocking effects or estimates of upslope and down-slope flows if observed values are entered into the analysis. Diagnostic procedures have modest computational requirements and can require fewer observations than does objective analysis to produce a three-dimensional wind field. Without representative data, however, diagnostic models cannot simulate such features as sea and land breezes. Prognostic numerical prediction models are intended to simulate all relevant physical processes without requiring a significant amount of observed data. These models require specification of the large-scale flow, surface conditions, and the initial state of the atmosphere. Because prognostic models simulate the temperature field in addition to the wind field, it is possible to determine atmospheric stabilities and mixing-height fields from the output. However, the computations performed by prognostic models can be expensive, and they do not necessarily reproduce available observations. Recent developments in data assimilation techniques could overcome the latter problem by forcing models to be more consistent with the available local observations, provided these observations can be shown to be truly representative of the actual meteorological field. Also, with better computer systems becoming available, it is increasingly practical to use full numerical prediction models. There are several hybrid models that use standard finite-difference techniques for horizontal advection but replace a rigid, vertical, finite-difference grid with one or more layers. The simplest example is a single-layer model in which the height and quantities of potential temperature and moisture, for example, are predicted in the boundary layer (Lavoie, 1972). This hybrid of numerical prediction and layer-averaged approaches is the basis of the EPA Regional Oxidant Model (ROM) (Lamb, 1983). An approach that is becoming more common is the use of the outputs from a prognostic model along with observed data as inputs to a diagnostic model. An EPA five-city study has identified important limitations in the routinely available meteorological data bases and raises several questions (Scheffe, 1990): Is it possible to set minimum data requirements given the great diversity of the areas in the United States to which air-quality models will be applied? Must data be of sufficient spatial and temporal density to result in

OCR for page 303
Page 310 such narrow uncertainty bounds that performance evaluation for wind field generation routines is not needed, or is "sufficient data" defined by the amount needed to attain a specified level of uncertainty? An emerging issue in wind field modeling is that of performance evaluation—determining a technical basis for judging the accuracy of simulated wind fields. Current criteria for evaluating wind field modeling performance have not been applied in wind field generation for ozone modeling. Performance evaluation of the meteorological model, independent of the air-quality model, is necessary to ensure that compensating errors are not introduced into predicted ozone concentrations through unjustifiable modifications of the wind field. There are several important meteorological variables other than wind field. Of particular importance is the treatment of photolysis rates and of the effects on these rates of clouds, urban aerosols, and ozone aloft. Clouds have traditionally been neglected in photochemical grid models because these models have focused on gas-phase pollutants, even though clouds can have a significant effect on the vertical distribution of pollutants (see Chapter 4) and on the attenuation (below cloud) or enhancement (near top of cloud) of photolysis rates. Objective or diagnostic techniques and prognostic modeling also can be used to calculate mixing heights. The key questions relate to the level of accuracy of the spatial and temporal variability in mixing heights. There are no currently accepted procedures for calculating mixing heights, and the mixing-height profile will strongly influence the predicted ozone concentrations in the modeling domain. Boundary And Initial Conditions When a grid-based photochemical model is applied to simulate a past pollution episode, it is necessary to specify the concentration fields of all the species computed by the model at the beginning of the simulation. These concentration fields are called the initial conditions. Throughout the simulation it is necessary to specify the species concentrations—called the boundary conditions—in the air entering the three-dimensional geographic domain. Three general approaches for specifying initial and boundary conditions for urban-scale applications can be identified: Use the output from a regional scale photochemical model; use objective or interpolative techniques with ambient observational data; or, for urban areas sufficiently isolated from significant upwind sources, use default regional background values and expand the area that is modeled and lengthen the simulation period to minimize uncertainties due to a lack of measurements. In the ideal case, observed data would provide information about the con-

OCR for page 303
Page 311 centrations at the model's boundaries. In practice, however, few useful data are generally available—a result of the difficulty in making measurements aloft and the fact that monitoring stations tend to be in places where air-quality standards are expected to be violated. An alternative approach is to use regional models to set boundary and initial conditions. This is, in fact, preferred when changes in these conditions are to be forecast. In any event, simulation studies should use boundaries that are far enough from the major source areas of the region that concentrations approaching regional values can be used for the upwind boundary conditions. Boundary conditions at the top of the area that is being modeled should use measurements taken from aloft whenever they are available. Regional background values are often used in lieu of measurements. An emerging technique for specifying boundary conditions is the use of a nested grid, in which concentrations from a larger, coarse grid are used as boundary conditions for a smaller, nested grid with freer resolution. This technique reduces computational requirements compared to those of a single-size, fine-resolution grid. Simulations of a multiday episode, beginning at night, when concentrations of ozone precursors are the lowest, minimize the influence of initial conditions on ozone concentrations predicted 2 and 3 days hence. Initial conditions are determined mainly with ambient measurements, either from routinely collected data or from special studies. Where spatial coverage with data is sparse, interpolation can be used to distribute the surface ambient measurements. Because few measurements of air-quality data are made aloft, it is generally assumed that species concentrations are initially uniform in the mixed layer and above it. To ensure that the initial conditions do not dominate the performance statistics, model performance should not be assessed until the effects of the initial conditions have been swept out of the grid. Demonstration of Attainment Chapter 3 introduced the SIP concept wherein the demonstration of attainment of the ozone NAAQS is based on future-year simulations. These demonstrations require projections of emissions and of initial and boundary conditions. The use of "background" boundary conditions would require an estimate of how these background conditions might change in response to regional changes in emissions. Boundary conditions can be based on future-year regional modeling. Initial conditions are typically reduced in proportion to emission reductions from the base to future year. A major question is "What is an acceptable procedure for demonstrating attainment?" Previous model applications for control strategy development

OCR for page 303
Page 312 have, for the most part, avoided complex scientific issues by focusing on one to three worst-case episodes, interpreting model results in a deterministic form without regard to modeling uncertainties and the statistical form of the ozone NAAQS. There is a critical need to identify and investigate methodologies for transforming results from deterministic models into a probabilistic form, so that informed decisions can be made about the efficacy of the selected emissions control strategy in achieving compliance with the ozone NAAQS. Other issues have not been adequately addressed in the context of ozone attainment demonstration. These include the number of episodes that need to be modeled and the duration of each episode. When the episodes to be modeled have been selected, it must be decided whether attainment should be demonstrated for all modeled episodes. For a particular episode, it is not a simple matter to decide which measure will be used to show attainment. For example, how should bias—under- or over-prediction of the peak ozone concentration—in the base year be addressed? One approach is to ignore any bias in the base year and consider attainment to be achieved if future-year predictions produce peak values below the NAAQS. Alternatively, the future-year peak ozone predictions could be normalized relative to the bias in the base-year simulation; if peak ozone is underpredicted by 10% in the base year, the future-year predicted peak ozone is increased by the same amount. Such questions related to how attainment is demonstrated have not been addressed adequately by the regulatory community. An aerometric data base is a critical component of a modeling application. Such a data base is needed to provide input to the model and to serve as a tool for assessing model performance. The elements of an aerometric data base for photochemical air-quality modeling are presented in Table 10-2. Most aerometric data bases use routine surface meteorological and air-quality measurements, either no upper-air meteorological data or routine National Weather Service balloon soundings only, and no air-quality data from aloft. A data base with such limitations does not contain enough information to characterize the three-dimensional meteorological and air-quality fields in space and time required for the model. An issue that must be addressed in an attainment demonstration is whether special aerometric measurements should be made over a limited period to provide a more extensive data base than is routinely available or whether a larger number of episodes should be considered with the more routine data base.

OCR for page 303
Page 313 Table 10-2 Aerometric Data Base Elements   Extent of database   Extensive/Intensive Routine Upper air meteorology (wind speed and direction, temperature, relative humidity, pressure)     Number of soundings per day 4-8 None or limited to routine National Weather Service or military installation observations Vertical resolution 100 meters None Air quality aloft     Species Ozone, NO, NO, speciated VOCs None Number of profiles per day per site > 3 None (Table continued on next page)

OCR for page 303
Page 340 values on second (or third) day concentrations, particularly in regions where the base-case predictions are highest. This simulation helps identify situations in which the base-case results are greatly affected by the boundary conditions. The simulation is performed by setting all inflow and outflow boundary values to zero, including those for the top surface of the modeling region. Zero Surface Deposition The zero-surface-deposition simulation addresses the influence of dry surface deposition removal on concentrations of primary and secondary species. The diagnostic run is performed by setting deposition velocities for all species to zero in the base-case simulation. Although deposition tests have not been reported in previous model evaluation studies, some general guidelines can be suggested. For primary species, such as NOx and VOCs, when deposition is neglected the downwind concentration fields should increase relative to the base case in a manner consistent with the deposition velocities for each primary species. For secondary species, such as ozone, the effect of a change in the deposition of primary species will depend on how that change propagates through the nonlinear chemistry. Mixing-Height Variations Mixing heights have a direct and often significant influence on ozone concentrations. The objective of the mixing-height diagnostic simulation is to reveal the degree to which ozone concentrations are influenced by the height of the mixed layer. At a minimum, one diagnostic run is suggested in which the hourly mixing-height values are uniformly increased by 50% above the base-case account. This increase is somewhat larger than the expected uncertainty in estimates of mixing heights typically encountered. Therefore, this simulation should provide a bound on the change in ozone predictions resulting from uncertainties in this input. Increased mixing heights typically reduce ozone concentrations, although the reduction is usually less than a one-to-one change. This effect can be seen by moving toward the origin along a line of constant VOC/NOx ratio on an ozone isopleth plot (see Chapter 6). One might choose, instead, to reduce the hourly mixing heights by 50%. The resultant changes in ozone concentrations under this scenario will typically be comparable in magnitude but of opposite sign to those for the mixing height increase.

OCR for page 303
Page 341 Dependence on Initial and Boundary Conditions Given the dependence of model results on initial and boundary conditions, it is of interest to quantify that dependence. Russell et al. (1989) performed a series of diagnostic simulations to identify how grid-based airshed model predictions change with changes in boundary and initial conditions. Initial conditions alone (in a simulation without emissions) led to the development of an air mass with elevated concentrations of ozone and VOCs that would last for several days before being slowly depleted by ground-level deposition and atmospheric chemistry. Predictions on the first and second day of a three-day simulation were dependent on initial conditions of VOC and NOx. By the third day, most of the initial-condition-dominated air mass had left the modeling region, although part of the region was still influenced. Boundary conditions primarily influence the regions near the boundary, with a small influence in the central modeling region. Reducing both boundary and initial conditions to background concentrations led to less than a 4% reduction in peak ozone and exposure predictions from the base-case simulation that used more representative values. However, if emissions are reduced by 50% or more, the role of initial and boundary conditions increases and cam become significant. For example, diagnostic simulations show that there are enough VOCs in the boundary conditions to form significant amounts of ozone when there are no VOC emissions from any source. Trajectory model studies show similar findings, and this is particularly important because trajectory model simulations are usually for less than 1 day. Simulations of less than 1 day are very sensitive to the choice of initial and upper level boundary conditions. Studies of boundary and initial condition effects suggest that large modeling domains and multiday (preferably, 3 or more) simulations are necessary for testing the effects of control strategies, and it is preferable to have model domain boundaries in relatively clean, rural regions. Assessing Simulation Results Decision makers and regulatory agencies seek quantitative performance standards by which to judge new models as acceptable. Each photochemical modeling episode exhibits distinctive aerometric and emissions features. Each model's available data base also is unique in the amount and quality of observations available to support model evaluation and testing. In addition, the particular set of modeling procedures and codes makes each application distinctive. Therefore, automatic use of standards for acceptance or rejection raises the risk of accepting a model evaluation that gives seemingly ''good" performance statistics but for

OCR for page 303
Page 342 the wrong or misleading reasons. It also could lead to rejection of a model evaluation that violates criteria for reasons related to input inaccuracies rather than to fundamental flaws. Instead of prescribing fixed performance standards, Tesche et al. (1990) suggest the following approach. From more than 15 years of photochemical model development and testing, photochemical grid model simulations generally produce peak (unpaired) prediction accuracy, overall bias, and gross error statistics in the approximate ranges of 15-20%, 5-15%, and 30-35%, respectively (Figure 10.8). A study that follows an approved ozone-modeling protocol but falls outside all these ranges would not be rejected unless evidence from the model's diagnostic simulations and the other numerical measures and diagnostic tests suggested unusual or aberrant behavior. For model simulations falling within these ranges, some additional diagnostic analyses could be appropriate to lend further support to the contention that the simulation is acceptable. For model results outside the ranges given for any one of these areas, it should be incumbent on the modeler to explain why the performance is poorer than that commonly achieved in similar applications and whether the causes of poorer performance will adversely affect the use of the model in control strategy evaluations. This method provides reviewing agencies with a general model performance target, but still guards against the inappropriate rejection of less accurate model simulations when appropriate explanations can be provided. Multispecies Comparisons The development of evaluation procedures that test photochemical model performance for species other than ozone can provide a basis for accepting or rejecting a model (or a model simulation); they significantly improve the chances that a flawed model will be identified. Adequate model performance for several reactive species increases the assurance that correct ozone predictions are not a result of chance or fortuitous cancellation of errors introduced by various assumptions. Multispecies comparisons could be the key in discriminating among alternative modeling approaches that provide similar predictions of ozone concentrations. To date, most model evaluation studies present results only for ozone (Tesche, 1988), although there are a few limited tabulations of NO2 predictions (Wagner and Ranzieri, 1984). Studies reported by Roth et al. (1983), Tesche (1983), Russell and Cass (1986), Wagner and Croes (1986), and Russell et al. (1988a,b) are among the few that present performance evaluation statistics for associated pollutants. Lack of ambient measurements for such pollutants is the major reason for the limited number of past studies. The data bases for the

OCR for page 303
Page 343 Southern California Air Quality Study and the San Joaquin Valley Air Quality Study (SJVAQS)/Atmospheric Utility Signatures, Predictions, and Experiments (AUSPEX) now allow for several comparisons with species other than ozone. The SCAQS data in particular afford a level of testing of photochemical models and modules, such as the chemistry mechanism, not previously possible. The availability of ambient air measurements for speciated organics and for key species such as formaldehyde (HCHO), peroxyacetyl nitrate (PAN), nitrogen dioxide (NO2), hydrogen peroxide (H2O2), nitric acid (HNO3), and organic acids will allow not only more extensive operational model testing but also diagnostic and comparative evaluations. Finally, the SCAQS data base, or similar data bases that will be assembled in the future, such as one from the SJVAQS/AUSPEX (Ranzieri and Thuillier, 1991), also offer the potential for evaluations of alternative chemical kinetic mechanisms. Evaluation of model performance for precursor and intermediate species as well as for product species other than ozone is recommended when ambient concentration data for these species are available. Comparisons of observed and predicted concentrations for all important precursors, intermediates, and products involved in photochemical air pollution—such as individual VOCs, nitric oxide (NO), nitrogen dioxide (NO2), PAN, ozone, H2O2, nitrous acid (HONO), and HNO3—are useful in model evaluation, especially with respect to the chemistry component of the model (Dodge, 1989, 1990). Accurate matching of ozone alone may not be sufficient to indicate that a chemical mechanism is correct. Comparisons of predictions and observations for total organic nitrates (mainly PAN) and inorganic nitrates (HNO3 and nitrate aerosol) can be used to test qualitatively whether the emissions inventory has the correct relative amounts of VOCs and NOx. However, HNO3 and nitrate aerosol cannot be included in the data set for model comparisons if the model does not include an adequate description of the HNO3 depletion process associated with aerosol formation. Depending on the availability of other measurements and the incorporation of aerosol dynamics and thermodynamic processes in the model, the above comparisons can be supplemented with others. For example, because HNO3 is a sink for the OH radical and because H2O2 is a sink for the HO2 radical, the ratio of HNOa to H2O2 is an indicator of the extent to which OH and HO2 radicals are adequately simulated (if nighttime HNO3 formation processes are properly accounted for). There are practical limitations in evaluating a model's performance in identifying speciated VOCs. Conceptually, comparisons can be made between observed and predicted total VOC concentrations as well as between observed and predicted concentrations of classes of VOCs. The second type of comparison requires aggregations of ambient VOCs into the classes used in the particular

OCR for page 303
Page 344 chemical mechanism of the model under consideration. However, emissions of nonreactive organic compounds, which can constitute 5 to 30% of actual VOC emissions, usually are not included in the simulation. Thus, predictions of VOCs could have an inherent bias toward underestimation relative to observed VOCs unless the difference is accounted for by explicitly excluding the nonreactive compounds from the observed concentrations. Also, approximations must be made in the schemes used to group the VOCs (Middleton et al., 1990). In some cases, the assignment of individual compounds to a class is based more on the similarity of their ozone formation potential to that of the model species than on the rate at which they react. Hence, perfect agreement is not expected and, in fact, agreement to within 20% for VOC classes is probably the best that can be expected. Mass Fluxes and Budgets Only recently have attempts been made to derive mass balances and carry out flux calculations for photochemical grid model simulations. This has occurred more routinely for regional grid-based models. Four mass balance and flux calculation procedures are suggested to accompany detailed performance evaluations. The first involves computing the mass fluxes into and out of the domain boundaries. The second procedure involves the mass fluxes into and out of the mixed layer. Third, the surface deposition fluxes should be estimated; hourly and daily average surface deposition rates should be calculated and reported for each species removed at the ground. In the final procedure, emissions, transport, transformation, and removal terms are reconciled in a simplified, closed mass budget over the whole modeling domain. The various flux terms described above, when combined with the hourly emissions rates, can be used in a simple mass budget to apportion the total mass in the modeling domain into emission, transport, and removal components. The transformation term is obtained by taking the difference between masses flowing into and out of the model domain, assuming a closed budget. Another test that should be performed to ensure mass consistency in meteorological and air-quality models is a simulation with an inert tracer version of the model. This simulation should be initialized with a homogeneous boundary and three-dimensional concentration field and should cover a period of at least 12 hours. The concentration field at the end of the simulation should be the same as the initial field.

OCR for page 303
Page 345 Sensitivity-Uncertainty Analysis Sensitivity analysis consists of systematically studying the behavior of a model over ranges in variation of inputs and parameters. This process can extend to studying the behavior of the model for changes in its basic structure—for different assumptions in its formulation. When model inputs and parameters are varied over their ranges of uncertainty to provide estimates of the range of uncertainty in predicted concentrations due to these input uncertainties, the process can be called sensitivity-uncertainty analysis. The diagnostic simulations discussed earlier fall within the general category of sensitivity analysis. Sensitivity analysis can be used to determine whether the predictive behavior of a model is consistent with what is expected on the basis of its underlying chemistry and physics—whether the model responds ''properly" when its inputs and parameters are varied. Sensitivity-uncertainty analysis is just a sensitivity analysis in which the variations in inputs and parameters correspond to their estimated uncertainties, and it is used to estimate the uncertainty in a model prediction. Sensitivity analysis of air-quality models meets two objectives: to determine qualitatively whether a model responds to changes in a manner consistent with what is understood about the basic physics and chemistry of the system, and to estimate quantitatively the uncertainty in model predictions that arise from uncertainties in the inputs and parameters. Various methods applicable to sensitivity-uncertainty analysis of photochemical air-quality models are available (Dunker, 1980, 1984; Seigneur et al., 1981; Tesche et al., 1981; Tilden et al., 1981; McRae et al., 1982; Brost, 1988; and Derwent and Hov, 1988). An overview and synopsis of major results of sensitivity testing and analyses of photochemical air-quality models can be found in Seinfeld (1988). There are several parameters of interest in the sensitivity analysis of photochemical air-quality models (Tesche et al., 1990): • Structure and design parameters of the model, including the horizontal and vertical dimensions of the computational grid cell, the number of cell layers in the vertical direction, and the size of integration time steps. In sensitivity testing, changes in these areas are deliberate and are related to model use. The objective is to identify values for each element that will lead to an optimal combination of computational efficiency and accuracy of prediction. • Constitutive parameters of the model, including chemical reaction rate constants and deposition velocities. Sensitivity analysis usually focuses on the effects on model predictions of uncertainty in these values. • Input parameters. These are calculated from the input data and, as discussed below, carry the uncertainties inherent in these data.

OCR for page 303
Page 346 When comparing model predictions and observations, one must remember that observations contain uncertainties due to measurement errors and the naturally random character of the atmosphere. Futhermore, models predict volume average concentrations, whereas observations reflect point measurements. Fox (1981) gives a good introduction to the concept of uncertainty in air-quality modeling. Beck (1987) offers further discussion of the concept of uncertainty in environmental models and data. The results of a sensitivity-uncertainty analysis can be presented graphically as indicated in Figure 10-7. The solid bold line in the figure represents base-case ozone predictions for particular air-monitoring stations. The boxes indicate the observed ozone concentrations at each hour, and the vertical lines associated with each box represent the estimated uncertainty of the ozone measurement. The magnitude of these measurement uncertainties has been estimated to include a component related to the spatial representativeness of the monitoring station. The solid lines enclose an ensemble of time series profiles obtained from several sensitivity runs involving different increases and decreases in the base-case mixing heights. In the example shown, these mixing-height uncertainties were derived from more than a dozen simulations of a numerical mixing-height model (Tesche et al., 1988c). Ideally, the ensemble of photochemical model predictions (enclosed by the thin solid lines) would trace a path within the upper and lower uncertainty bounds of the hourly ozone measurements. Testing the Adequacy of Model Response to Changes In Emissions It is important to assess the ability of models to correctly simulate the effects of emissions changes because of the direct connection between changes in emissions and the intended regulatory application of photochemical models. Traditionally, photochemical models are evaluated for a variety of meteorological conditions over periods of time too brief to involve major changes in emissions. Then the critical assumption is implicitly made that the models will be applicable under conditions of drastically altered emissions. The work of Dennis and co-workers (Dennis et al., 1983; Dennis and Downton, 1984; Downton and Dennis, 1985; Dennis, 1986) has shown that grid-based photochemical models that perform adequately for a range of meteorological conditions do not necessarily work well when the evaluation involves a large change in emissions. They have found that different versions of the UAM that give similar performance results under conditions of changing meteorology perform very differently when tested for emissions changes. (These versions of the UAM represent progressive improvements in chemistry, numerical methods, and the treatment of meteorolo-

OCR for page 303
Page 347 gy.) It is imperative, therefore, to evaluate photochemical models intended for use in the development of air pollution control strategies to determine their ability to simulate the effects of emissions changes. The problems raised in this kind of evaluation are serious. To allow a meaningful model performance evaluation, detailed emissions inventories of comparable accuracy are required for base years long enough apart (at least 10 years) that major emissions changes have taken place. It also is necessary to identify episodes that occur in periods of similar meteorology for the two representative years. Ideally, this kind of evaluation, which uses historical emissions and air-quality records, is the preferable one, but generally the lack of detailed inventories and historical aerometric data prohibits this approach. Even if the required data are available, the evaluation would have to account for all the changes that have occurred over the years in the procedures for developing emissions inventories, monitoring air quality, and so on. The use of weekday versus weekend emission rates has been suggested as an alternative to retrospective modeling. Even assuming that inventories are accurately estimated, it is not likely that the difference in emissions would be sufficient for a meaningful evaluation of model performance, although such studies would be valuable. An exception might be to conduct such an evaluation after the implementation of a major regulation such as RVP (Reid vapor pressure) reduction of motor vehicle fuel. A third promising approach to evaluating a model's ability to correctly predict the effects of major changes in emissions would be to thoroughly test the model for different urban areas, using input data sets of similar levels of detail. Because evaluation using historical inventories and aerometric data does not appear feasible now, this third option appears to be a good alternative. A fixed version of the photochemical model (same horizontal and vertical resolution; identical input data preprocessors, chemistry, and removal modules; and so on) could be applied to all regions selected for the evaluation. The evaluation could span a wide range of meteorological conditions for the urban areas under consideration, corresponding to high-, moderate-, and low-ozone days. Such an evaluation would not test uniform changes in emission; instead, it would evaluate overall model performance for different spatial and temporal distributions, source strengths, and speciation of emissions. Even if the problems of availability and quality of input data (emissions and aerometric) are solved, allowing one to evaluate a model's ability to simulate significant emissions changes, one must still account for the fact that the sensitivity of a photochemical modeling simulation to emissions changes will vary according to meteorology. Wagner and Wheeler (1989), reporting on sets of simulations performed by Tesche et al. (1988a,b) and Wagner (1988), concluded that "the location and amount of maximum sensitivity to emissions changes vary

OCR for page 303
Page 348 with the meteorology. This may mean that more than one episode should be used in evaluating the effects of emission changes upon peak ozone concentrations." Indeed, the selection of particular ozone episodes on which to design emissions controls can have a substantial effect on the projected control levels. It is therefore important to examine several episodes to determine the sensitivity of controls to meteorology. Summary There are several classes of photochemical air-quality models (Table 10-5), and although grid-based models are the best for simulating atmospheric chemistry and transport processes, they require relatively large data bases, and many areas of the country do not have the resources to support their use. These areas will continue to rely on trajectory models like EKMA (empirical kinetic modeling approach) to determine ozone abatement strategies. Two questions attend the use of EKMA-type models and reduce the confidence that can be placed in guidance derived from them: First, what inaccuracies result if vertical heterogeneity is not included in the trajectory model? Second, to what extent is a trajectory model simulation an adequate representation of three dimensional airshed behavior? Some results suggest that EKMA-type trajectory models are too limited in their formulation to account for multiple-day episodes of high ozone concentrations. For example, application of EKMA-type models to Boston (Chang et al., 1989) and Philadelphia (Whitten et al., 1986) showed either little change or an increase in ozone concentrations in response to NOx reductions. However, multiple-day ROM simulations (Possiel et al., 1990) found that NOx reductions led to decreased ozone concentrations in both locations. The model inputs needed to simulate historical ozone episodes—boundary and initial conditions, both on the ground and aloft, and emissions inventories—have associated uncertainties, often of a magnitude difficult to estimate. Compilation of simulations over many episodes and many regions indicates a general under-prediction of peak ozone concentrations. The most consistent explanation for this behavior is a general underestimation of volatile organic compound (VOC) emissions. In some cases, however, the uncertainties in model inputs are large enough that the temporal and spatial features of ozone behavior can be reproduced by selection of the inputs within their ranges of uncertainty. Indeed, the "play" in inputs that has been used to improve ozone predictions might have compensated for an underestimated VOC emissions inventory.

OCR for page 303
Page 349 TABLE 10-5 Classes of Photochemical Models Model Type Strengths Limitations EKMA (Empirical kinetic modeling approach) Easy to apply, detailed chemistry, computationally rapid. Lacks physical detail, short model simulations. Does not accurately simulate multi-day events or long-range transport. Urban, grid-based Physically detailed. Suited for multiday modeling of urban areas (˜400 km). Computationally demanding. Sensitive to boundary conditions when long-range transport is important. Regional Physically detailed. Suited for studying regional areas (˜1000 km). Computationally demanding, limited spatial resolution. (ROM also has limited vertical resolution.) Not well suited to studying pollutant dynamics in cities. Nesteda Advantages of both urban and regional models. Computationally awkward and demanding. Information travels in one direction. Multiscaleb Advantages of nested, urban, and regional models. Computationally straightforward. Computationally demanding. In development. aNested models are models in which finer grid scales are embedded. bMultiscale models are in effect nested models whose character on different scales may be different. If so, this is a cause for concern when models are then used to determine degrees of VOC and NOx control needed to attain the ozone National Ambient

OCR for page 303
Page 350 Air Quality Standard (NAAQS). The critical question is "What is the effect of uncertainties in base-year inputs on projections of control levels for future years?" Once this question is answered, these uncertainties need to be incorporated in the State Implementation Plan. For example, it is important to know whether such uncertainties affect the choice of control strategies—for example, control of VOCs versus control of oxides of nitrogen (NOx) versus both. Model performance evaluation procedures must be designed to reveal flaws in a base-case simulation to ensure that a model gives the right answer for the right reason. Computational constraints historically have limited the use of advanced three-dimensional, photochemical air-quality models. Instead, less-comprehensive, less computationally intensive, and more limited models, such as EKMA have been used. Such models are not capable of fully characterizing ozone dynamics in urban and regional areas over multiple days, nor the response of ozone to emission changes. Rapid increases in computational power and algorithmic efficiencies now allow for more widespread use of advanced models, and are playing a significant role in the ability to understand atmospheric pollutant dynamics. The continued evolution of computational capability will allow for in-depth studies using more chemically and physically comprehensive models.