AEROSOL-FORMING POTENTIAL OF ATMOSPHERIC ORGANIC COMPOUNDS

Transcription

1 AEROSOL-FORMING POTENTIAL OF ATMOSPHERIC ORGANIC COMPOUNDS John H. Seinfeld 1, Prasad Pai 2, and David Allen 3 1 California Institute of Technology 2 AER, San Ramon, CA The University of Texas, Austin, TX CONTENTS CONTENTS... 1 INTRODUCTION... 1 THEORY OF SECONDARY ORGANIC AEROSOL FORMATION... 2 EXPERIMENTAL DETERMINATION OF SOA YIELDS... 3 TREATMENT OF SECONDARY ORGANIC AEROSOLS IN CURRENT ATMOSPHERIC MODELS... 7 INSIGHTS AVAILABLE FROM DATA ON AMBIENT ORGANIC AEROSOL... 9 REFERENCES INTRODUCTION Urban fine particulate matter is comprised of a complex mixture of both primary and secondary organic and inorganic compounds and emanates from a wide variety of sources. An important component that can significantly contribute to the fine particulate burden, especially during severe urban smog episodes, is secondary organic aerosol (SOA). Like ozone, secondary organic aerosol results from the atmospheric oxidation of reactive organic gases (ROGs), but whereas the oxidation of most ROGs results in ozone formation, SOA is generally formed only from the oxidation of ROGs comprised of six or more carbon atoms. This is because oxidation products must have vapor pressures that are sufficiently low to enable them to partition into the aerosol phase. The atmospheric chemical reaction pathways of ROG molecules sufficiently large to lead to SOA are complex, and resulting oxidation products are both numerous and difficult to quantify analytically. As a result, it is currently not possible to determine the aerosol formation potential of individual ROGs and their contribution to the secondary organic urban particulate burden strictly on the basis of atmospheric chemical reaction mechanisms. The chemical process of organic aerosol formation involves the formation of semivolatile organic gases, S 1, S 2,..., from the gas-phase reaction of a parent hydrocarbon, HC, with the OH radical, 1

2 k OH HC + OH... + α 1 S 1 + α 2 S (1a) where k OH is the OH reaction rate constant, and α 1, α 2,...are the stoichiometric product coefficients. Later these stoichiometric coefficients will be expressed on a mass basis, rather than the usual molar basis. If the parent hydrocarbon is an alkene, reactions with O 3 and NO 3 radicals are also possible, providing additional pathways for semi-volatile product formation, ko3 HC + O α 1,O 3 S 1,O3 + α 2,O3 S 2,O (1b) kno3 HC + NO α 1,NO 3 S 1,NO3 +α 2,NO3 S 2,NO (1c) The first-generation products, S 1, S 2,... may subsequently undergo gas-phase reaction themselves creating second-generation condensable products, S 1a, S 1b,... and S 2a, S 2b, etc., k OH,S1 S 1 + OH... +α 1a S 1a +α 1b S 1b +... k OH,S2 S 2 + OH... +α 2a S 2a +α 2b S 2b +... (2a) (2b) where k OH,S 1 and k OH,S2 are the OH-reaction rate constants for the products, S and S, 1 2 respectively. Secondary organic aerosol yields have been measured for many individual ROGs by a number of researchers over the last decade or more (Izumi and Fukuyama, 1990; Wang et al., 1991ab; Pandis et al., 1991; Zhang et al., 1991; Odum et al., 1996, 1997ab; Hoffmann et al., 1997; Forstner et al., 1997ab; Griffin et al., 1998). Initially it was believed that each ROG should possess a unique value of its SOA yield (Grosjean and Seinfeld, 1989; Pandis et al., 1992, 1993), but measured yields for an individual ROG exhibited a wide degree of variation that could not be reconciled in terms of a single, unique SOA yield for each parent ROG. Following Pankow (1994ab), Odum et al. (1996) formulated a framework for explaining observed SOA yield data. They suggested that secondary organic aerosol formation is best described by a gas/aerosol absorptive partitioning model. Within that framework, semi-volatile products from the atmospheric oxidation of an ROG can partition into an absorbing organic aerosol (om) phase at a concentration below their saturation concentration, analogous to the partitioning that occurs between the gas and aqueous phases of a water-soluble atmospheric constituent. THEORY OF SECONDARY ORGANIC AEROSOL FORMATION Equilibrium gas-particle absorptive partitioning of a semi-volatile organic species i between the gas phase and an organic phase is described by (Bowman et al., 1997) 0 p i = x i ζ i p i (3) 2

3 where p i (torr) is the gas-phase partial pressure of species i, x i is the mole fraction of species i in the aerosol phase, ζ i is the activity coefficient of species i in the aerosol-phase organic mixture, and p i o (torr) is the vapor pressure of species i as a pure liquid (subcooled, if necessary). The gas-phase partial pressure, p i, can be converted to the gas-phase mass concentration, G i (µg m 3 ), by the relationship G i = p i m w RT 106 (4) where m w (g mol 1 ) is the mean molecular weight of the absorbing organic matter, R (=6.2 x 10 2 torr m 3 mol 1 K 1 = J mol 1 K 1 ) is the ideal gas constant and T (K) is temperature. The factor 10 6 accomplishes the appropriate unit conversions. For organic species with similar molecular weights, the aerosol mole fraction, x i,, is given by x i = k A i A k + M init (5) where A i (µg m 3 ) is the aerosol mass concentration of species i, Σ A k, (µg m 3 ) is the total aerosol mass concentration of all the individual semi-volatile organic species, and M init represents any initially present absorbing organic mass. Equations (4) and (5) can be substituted into equation (3) and rearranged to yield A i G i ( A k + M init ) = k RT m w ζ i p i = K i (6) where K i (m 3 µg 1 ) is defined as the absorption partitioning coefficient of species i (Pankow, 1994a, b; Odum et al., 1996). The absorption partitioning coefficient incorporates vapor pressure, activity coefficient, and molecular weight, providing a single equilibrium parameter for each compound. K i is analogous to a Henry s law coefficient in relating gas-phase concentrations of species i to the mass fraction of species i in the aerosol phase. An important implication of Equation (6) is that, since, at a particular temperature, K i is a constant, a greater fraction of each product must partition to the organic phase as the total organic aerosol concentration increases. EXPERIMENTAL DETERMINATION OF SOA YIELDS The approach that has been most successful to estimate the SOA forming capability of a ROG involves the direct measurement of secondary organic aerosol yields. The SOA yield Y is a measure of the mass of aerosol that is produced from the atmospheric oxidation of an ROG and is defined as Y = M o ROG (7) 3

4 where M o is the amount of aerosol produced (µg m 3 ) for a given reacted amount of an ROG, ROG (µg m 3 ). In the study of secondary organic aerosol formation, typically a smog chamber is initially filled with a mixture of NO x, inorganic seed particles, and an aerosol-producing hydrocarbon. The chamber is then exposed to sunlight, or other UV sources, that initiates photooxidation. As the hydrocarbon reacts it forms semi-volatile products that condense on the seed particles. If mass transport to the available particles cannot keep up with the rate of product formation or when a seed aerosol is not initially present, the semi-volatile products accumulate in the gas phase until supersaturation is reached, and nucleation occurs (Bowman et al., 1997). Studies show that the amount of aerosol produced for a given amount of reacted ROG is independent of whether a seed aerosol is present or not. Reactions are ordinarily run until the entire initial amount of ROG is consumed. Typically the volume of the initial seed aerosol is small compared to the organic aerosol volume generated, and A = M k k o The total SOA yield from an ROG that generates N semi-volatile products is computed as follows. The total concentration of product i, C i, is proportional to the total amount of parent organic that reacts, ROG, α i ROG = C i (8) C i is also equal at any time to the sum of the gas- (G i ) and aerosol - (A i ) phase concentrations of i, C i = A i + G i (9) The yield of product i, Y i, is A i / ROG. The total SOA yield is just the sum of the individual product yields, N Y = Y i (10) i=1 Combining these relations with the definition of the gas-particle partitioning constant, K i, gives the expression for the total yield in terms of the individual product stoichiometric coefficients α i and partitioning coefficients, K i, and the organic aerosol mass concentration, M o, N Y = M o i=1 Note that in the limit of small organic aerosol mass concentration, α i K i 1+ K i M o (11) N Y ~ M o α i K i (12) i=1 the SOA yield is directly proportional to the amount of organic aerosol mass M o, and that in the limit of large M o, 4

5 N Y ~ α i (13) i =1 where the total yield is independent of M o and is just the sum of the mass-based stoichiometric coefficients of the products. It is important to distinguish between aerosol yield, Y, and stoichiometric coefficients, α i. Stoichiometric coefficients depend on the gas-phase chemical mechanism and are assumed constant; they represent the total amount of semi-volatile product formed, in both gas and aerosol phases, per amount of parent hydrocarbon reacted. The yield, on the other hand, which measures only the semi-volatile products that have partitioned into the aerosol phase, is not constant but will vary depending on the amount of organic mass available as an absorption medium. Stoichiometric coefficients by themselves are therefore not sufficient to predict the amount of aerosol formation (except in the limit of very large organic mass, as in equation (13)). Partitioning coefficients, stoichiometric coefficients, and organic aerosol mass are required in general to determine the SOA yield. Ozone-forming potential of organics is determined based on atmospheric reaction mechanisms. In principle, aerosol-forming potential could be calculated based on a similar atmospheric oxidation mechanism that includes all significant semi-volatile product species. The actual amount of aerosol that would be formed under a particular set of circumstances, unlike ozone, depends on the amount of aerosol available to absorb the semi-volatile products. The relative aerosol-forming potential of a group of organics could, in principle, be determined based on their oxidation products and the thermodynamic properties of these products. This ab initio approach represents a goal that is not yet attainable because of incomplete knowledge of the semi-volatile oxidation products of the important aerosol-forming compounds. Thus, it is necessary to rely on experimentally measured aerosol yields. Aerosol yields Y are expressed theoritically as a fraction of the available aerosol organic mass M o through Equation (11). Observed aerosol yields as a function of M o can be fit to Equation (11) by specifying the stoichiometric coefficients α i and gas-particle partitioning coefficients K i of each of the semivolatile products of oxidation of the parent organic. Over the last several years, SOA yields for over 30 aromatic and biogenic organics have been measured in the California Institute of Technology outdoor smog chamber (Table 1). In order to fit the observed yields to Equation (11), the mix of semi-volatile oxidation products for each parent compound has been represented by two empirical products, characterized by parameters α 1, K 1, and α 2, K 2. It has been determined that observed yields cannot be fit by assuming only a single product and that three products is superfluous (Odum et al., 1996). Roughly speaking, one of the empirical semi-volatile products tends to represent a relatively lower vapor pressure compound and the other a relatively higher vapor pressure compound. Low-yield aromatics represent those species that fall on the lower of two curves of Y versus M o described by Odum et al., (1997ab). Correspondingly, high-yield aromatics represent the species falling on the higher of the two curves. 5

6 Table 1. Aerosol Formation Parameters α i and K om,i (m 3 µg -1 ) Obtained in the Caltech Smog Chamber (Odum et al., 1996, 1997a, 1997b; Hoffmann et al., 1997; Griffin et al., 1998) Hydrocarbon α 1 K 1 α 2 K 2 Aromatic Compounds (17) Low-yield aromatics High-yield aromatics Diethylbenzene Methylpropylbenzene Biogenic Compounds (13) 3 -Carene β-caryophyllene α-humulene Limonene Linalool Ocimene α-pinene β-pinene Sabinene α-&γ-terpinene Terpinene-4-ol Terpinolene It must be noted that this theory assumes that secondary products are unable to form a solution with existing inorganic seed aerosol. Accounting for the interactions between the organic compounds themselves allows it to be shown that such products can condense onto seed aerosol at concentrations lower than those predicted by saturation theory alone (Seinfeld and Pandis, 1998). In this case, the threshold amount of parent compound that must react to form secondary organic aerosol is defined as ROG*. After consumption of ROG*, products condense onto seed aerosol to form an initial organic layer that can then act as an absorptive medium. At this point, absorption becomes the dominant mechanism governing the partitioning of secondary products and, therefore, determining yield, as in the atmosphere. 6

7 The stoichiometric and partitioning parameters allow direct evaluation of the aerosolforming potential of the parent organics. The SOA yield Y is the mass of aerosol produced per unit mass of parent organic reacted, and Y depends on the amount of organic aerosol mass available to absorb the semi-volatile oxidation products. SOA potentials are given by Odum et al. (1997ab) for 17 aromatic precursors and by Griffin et al. (1998) for 14 biogenic precursors. Odum et al. (1997ab) showed, moreover, that aerosol yields for the photooxidation of a mixture of parent hydrocarbons can be predicted simply as the sum of the SOA yields for the individual parent compounds. This suggests that, at least for the case of a pure organic absorbing phase, oxidation products of different parent hydrocarbons are as soluble in a mixed organic product phase as in an organic phase consisting exclusively of their own oxidation products. The experimentally determined SOA yields reported by Odum et al. (1997ab) and Griffin et al. (1998) have been measured at relative humidity (RH) less than 5%. At this level of RH the seed aerosol, (NH 4 ) 2 SO 4, is dry and the resulting organic aerosol is water-free. Experiments are presently underway at Caltech to measure SOA yields as a fraction of RH over realistic ambient RH ranges. Because organic products will likely be most soluble in their own liquids, SOA yields measured at essentially 0% RH can be expected to represent an upper limit to the aerosol partitioning that will result. While many SOA products are water soluble (Saxena and Hildemann, 1996), they are not expected to be more soluble in an aqueous mixture then in a pure organic phase. TREATMENT OF SECONDARY ORGANIC AEROSOLS IN CURRENT ATMOSPHERIC MODELS Because of the difficulties in characterizing SOA on a molecular basis and considering the significant uncertainties associated with model input data - for example emissions of VOC that are SOA precursors - the treatment of SOA in current air quality models is limited. Seigneur et al. (1998) recently reviewed the treatment of SOA formation in current regulatory and research-grade air quality models. Table 2, adapted from Seigneur et al. (1998), compares the SOA treatment in eight current air quality models. Details on the individual air quality models can be obtained from the original references provided in Table 2. Of the eight models that treat SOA formation, six (DAQM, RPM, SAQM-AERO, UAM- AERO, UAM-AIM and VISHWA) use the lumped SOA yield approach of Pandis et al. (1992), which makes use of the VOC-specific fractional aerosol coefficients of Grosjean and Seinfeld (1989). In this approach, each class of VOC in the gas-phase chemical kinetic mechanism is assumed to lead to a fixed fraction of SOA product through its oxidation reactions. This results from the basic assumption that a fixed fraction of the products of the VOC oxidation is condensable. The amount of the condensable organics that exceeds the corresponding saturation vapor pressure is converted to the particulate phase. The saturation vapor pressure in most models is typically set to zero or a small value (1ppt). The six air quality models that employ the lumped SOA yield approach of Pandis et al. (1992) use different gas-phase mechanisms. These 7

8 Table 2. Treatment of SOA Formation in Current Air Quality Models Air Quality Model Secondary Organic Aerosol Treatment Gas Phase Chemical SOA Formation Reference Mechanism CIT Extended LCC Absorption mechanism with mass transfer Meng et al. (1998) DAQM RADM2 Lumped aerosol yields of Pandis et al. (1992) Moucheron and Milford (1996) GATOR GATOR Solubility in water Jacobson (1997) REMSAD Condensed CBM-IV chemistry VOC emission fraction Guthrie et al. (1995) RPM RADM2 Lumped aerosol yields of Pandis et al. (1992) Binkowski and Shankar (1995) SAQM-AERO CBM-IV and SAPRC Lumped aerosol yields of Pandis et al. (1992) Dabdub et al. (1998) UAM-AERO CBM-IV and SAPRC Lumped aerosol yields of Pandis et al. (1992) Lurmann et al. (1997) UAM-AIM CBM-IV and SAPRC Lumped aerosol yields of Pandis et al. (1992) Sun and Wexler (1998) UAM-LC Parameterized chemistry None Lurmann and Kumar (1996) VISHWA Condensed chemistry (7 reactions and 7 species) 2-component approach with aromatic and terpene lumped aerosol yields of Pandis et al. (1992) adjusted downward to match ambient observations Venkatram et al. (1997) gas-phase mechanisms include the Carbon Bond Mechanism (CBM-IV), RADM2, GATOR, SAPRC, and extended LCC mechanisms, which are described by Gery et al. (1989), Stockwell et al. (1990), Jacobson (1994), Carter (1990, 1995), and Harley et al. (1993), respectively. In five of the six air quality models listed in Table 2, the lumped SOA yields of Pandis et al. have been used without any modifications. In the VISHWA model, the lumped SOA yields of Pandis et al. for the two lumped SOA precursors used, i.e., aromatics and terpenes, have been adjusted downward by a factor of about eight, reportedly to match experimental observations. UAM-LC does not currently treat SOA formation and REMSAD uses fixed yields for VOC emissions, i.e., it does not have kinetic treatment of SOA formation. GATOR does not use the lumped SOA yield approach but instead describes gas-particle partitioning of about ten condensable and soluble organic species that, however, are not directly representative of ambient condensable reaction products. In CIT, SOA is modeled by mass transport between gas and particulate phases, which is governed by the same formulation as that for inorganic volatile compounds. The surface vapor concentrations of semi-volatile organic compounds G i eq are calculated according to Equation (6), G i eq = A i K i OM (14) 8

9 where A i (µg m -3 ) is the concentration of organic species i in the aerosol phase, K i (m 3 µg -3 ) is the absorption partitioning coefficient, and OM(µg m -3 ) is the absorbing organic mass concentration. Recent work by Strader et al. (1998) combines the organic-phase absorption approach of Odum et al. (1996, 1997ab) with recent experimental data on saturation vapor pressures of organic compounds. Strader et al. (1998) used a model with six condensable products (6 partition parameters): one from alkane precursors (also for benzaldehyde phenol, cresol, and nitrophenol), three from aromatic precursors, one from alkene precursors, and two from monoterpenes. Each precursor forms one or two condensable products, resulting in 14 yield parameters determined on the basis of data from the Caltech chamber. This approach has been incorporated in the revised version of the UAM-AERO air quality model. The key differences between Odum et al. and the Strader et al. approaches lie in the determination of the absorption partitioning coefficient K i (m 3 µg -3 ) and in the precursor/product species that are modeled. As noted earlier, Odum et al. (1997ab) assumed two condensable products for each aerosol precursor, with the absorption model fitted to data obtained in smog chamber experiments to determine empirically the yield and partition parameters. Strader et al.,(1998) invoked solution theory in their determination of the partition coefficients. Raoult s law describes the equilibrium partition of an ideal gas-ideal solution system, y i p = x i p i sat where y i p is the partial pressure of compound i in the gas phase, x i is the mole fraction of i in sat the liquid phase, and p i is the saturation vapor pressure of i at the system temperature. Converting the units from pressure to mass concentration in the gas phase, c i gas = x i c i gas.sat Instead of the equilibrium constant K i, the partitioning of compound i is calculated via the saturation concentration, c gas.sat i, of i in the gas phase. Since the identities of the condensing SOA gas.sat compounds are frequently unknown, c i is chosen based on smog chamber results and laboratory experiments for the temperature dependence of the saturation concentrations. The methodology of Strader et al. (1998) provides the framework for incorporating mechanistic information of the condensing species when it becomes available. At present, however, both methods rely on empirical data to determine the partition parameters. (15) (16) INSIGHTS AVAILABLE FROM DATA ON AMBIENT ORGANIC AEROSOL In the interest of space, we do not endeavor to survey here data on ambient levels of secondary organic aerosol in different areas of the United States. We refer the reader to the recent review of Turpin et al. (1999) and references therein. The most extensive data base on secondary organic aerosol is available for the California South Coast Air Basin, where levels can 9

10 reach 50% or more of the organic PM 2.5 mass during photochemical pollution episodes. The extent of importance in other areas of the country will need to be evaluated through a combination of ambient measurements and atmospheric modeling. Data on the composition of ambient organic aerosol can provide some insights into the magnitude of secondary organic aerosol formation and the chemical pathways and precursors that are responsible for the aerosol formation. Three types of data on ambient aerosol can be used in these assessments: Data on organic and elemental carbon in ambient aerosol can be used to estimate the overall magnitude of secondary aerosol formation; Data on organic compound classes, such as carbonyl groups and organonitrate groups, can be used to assess the magnitude of secondary organic aerosol formation and the approximate composition of the secondary aerosol; Concentrations of molecular tracers can be used to assess the role of individual hydrocarbon precursors in secondary organic aerosol formation. Each of these approaches is discussed briefly below. Often, the only data available on the carbonaceous fraction of organic aerosol are measurements of organic and elemental carbon. Elemental carbon is roughly defined as the material that will not thermally desorb from a filter sample and is generally attributed to graphitic, soot-like structures. Elemental carbon is assumed to be exclusively due to primary emissions. Organic carbon is roughly defined as material that will thermally desorb from a filter sample and may be associated with either primary or secondary aerosol. The most common approach for using ambient data on organic and elemental carbon to estimate the magnitude of secondary organic aerosol formation involves using the ratios of elemental to organic carbon. (Turpin and Huntzicker, 1991; Turpin et al., 1991; McMurry, 1989; McMurry and Zhang, 1989; Gray et al., 1986) By assuming that all of the elemental carbon is primary and that the primary organic aerosol has a characteristic ratio of organic to elemental carbon, it is possible to estimate the fraction of organic aerosol that is primary. The remainder of the organic aerosol is assumed to be secondary. The accuracy of these estimates of secondary organic aerosol formation is highly dependent on having accurate values for the ratio of organic to elemental carbon in the primary aerosol - a quantity that is not known with certainty. Thus, estimates of secondary aerosol formation based on organic and elemental carbon measurements may be subject to inaccuracies. Nevertheless, because organic and elemental carbon measurements are often the only data available on the carbonaceous component of atmospheric aerosol, these methods will remain in common use. Functional group measurements are an alternative to using organic and elemental carbon to estimate secondary organic aerosol concentrations. A large fraction of organic aerosol can be characterized using compound class or functional group measurements based on infrared spectroscopy (Pickle et al., 1990; Mylonas et al., 1991; Allen et al., 1994). In these functional group measurements, the organic fraction of an ambient sample is typically assumed to consist of aliphatic carbon, aromatic carbon, carbonyl, organitrate, nitroaromatic, and a few other 10

11 functional groups. Some of these functional groups, such as ketones, aldehydes and organonitrates, are due almost exclusively to secondary organics. Other functional groups, such as aliphatic and aromatic carbons could be either primary of secondary. Thus, the concentrations of groups like ketones, aldehydes and organonitrates can provide a lower bound on secondary aerosol formation and assumptions about average structures of the compounds that make up secondary organic aerosol can be used to estimate total secondary organic concentrations. Again, this method of estimating secondary organic aerosol has uncertainties associated with it. The method does provide, however, measurements that are complementary to the usual measurements of elemental and organic carbon, and the functional group measurements begin to provide some insight into the chemical structures making up the secondary organic aerosol. Neither organic carbon/elemental carbon measurements nor functional group measurements identify which gas phase precursors led to the secondary aerosol. In order to attribute secondary organic aerosol to specific hydrocarbons precursors, it is necessary to perform molecular level characterizations of ambient organic aerosol. A number of recent studies (see, for example, Hildemann, et al., 1994) have been used individual compounds as tracers for aerosol sources. This molecular tracer approach has been quite successful in identifying the major sources of primary carbonaceous aerosol, but to date, the method has not been applied to secondary aerosol because the yields and stabilities of the molecular products of gas-phase photooxidation reactions are relatively unknown. A literature on molecular tracers for secondary organic aerosol is beginning to emerge, however, and this approach to understanding the precursors for secondary aerosol is likely to become increasingly important. To summarize, ambient data on aerosol composition provides some information on the magnitude and structure of secondary organic aerosol, but this information is incomplete and requires that significant assumptions be made in interpreting the ambient data. Our understanding of the chemical nature and sources of secondary organics would be significantly improved by the identification of suitable molecular tracers of organic aerosol formation pathways. Atmospheric chemical reaction mechanisms currently included in urban-and regionalscale models were designed for accurate prediction of ozone formation chemistry. Generally, these mechanisms do not account for the chemistry of organics in sufficient detail to predict the generation of semi-volatile products. Many of the higher molecular weight organics that are not important in ozone formation, but are sources of secondary organic aerosol, are not represented in current mechanisms. To predict SOA formation, chemical reaction mechanisms need to be expanded to include all important SOA-forming organics and to include, to the extent known, a representation of the semi-volatile products formed from each parent compound. Evaluation of the organics that need to be included will be based on available chamber data, on estimated atmospheric oxidation mechanisms, and on the estimated vapor pressures of the oxidation products. While this expansion may not lead to substantially improved ozone prediction, it is essential to prediction of SOA formation. 11

12 REFERENCES Allen, D. T., Palen, E. J., Haimov, M. I., Hering, S. V., and Young, J. R. (1994) Frourier transform infrared spectroscopy of aerosol collected in a low pressure impactor: Method development and field calibration, Aerosol Science and Technology, 21, Binkowski, F. S. and Shankar, U. (1995) The regional particulate matter model, 1: model description and preliminary results, J. Geophys. Res., 100, 26,191-26,209. Bowman, F.M., Odum, J. R., Pandis, S. N., and Seinfeld, J. H. (1997), Mathematical model for gas/particle partitioning of secondary organic aerosols, Atmos. Environ., 31, Carter, W. P. L. (1990) A detailed mechanism for the gas phase atmospheric reactions of organic compounds, Atmos. Environ., 24A, Dabdub, D., Dehaan, L. L., Kumar, N., Lurmann, F., and Seinfeld, J. H. (1998) Computationally efficient acid deposition model for California, Draft Report, ARB Contract #92-304, California Resources Board, Sacramento, California. Forstner, H. J. L., Flagan, R. C., and Seinfeld, J. H. (1997a) Secondary organic aerosol formation from the photooxidation of aromatic hydrocarbons: molecular composition. Environ. Sci. Technol., 31, Forstner, H. J. L., Seinfeld, J. H., and Flagan, R. C. (1997b), Molecular speciation of secondary organic aerosol from the higher alkenes: 1-octene and 1-decene, Atmos. Environ., 31, (1997). Gery, M. W., Whitten, G. Z., Killus, J. P., and Dodge, M. C. (1989) A photochemical kinetics mechanism for urban and regional scale computer modeling, J. Geophys. Res., 94. Gray, H. A., Cass, G. R., Huntzicker, J. J., Heyerdalh, E. K., and Rau, J. A. (1986) Characteristics of atmospheric organic and elemental carbon particle concentrations in Los Angeles, Environ. Sci. Technol., 20, Griffin, R. J., Cocker, D. R., Flagan, R. C., and Seinfeld, J. H. (1998) Organic aerosol formation from the oxidation of biogenic hydrocarbons, J. Geophys. Res., in press. Grosjean, D. and Seinfeld, J. H. (1989) Parameterization of the formation potential of secondary organic aerosols. Atmos. Environ,. 23, Guthrie, P. D., et al. (1995) Development and preliminary testing of the regulatory modeling system for aerosols and depostion (REMSAD), technical memorandum SYSAPP-96111, Systems Applications International, San Rafael, California. 12

13 Harley, R. A., Russell, A. G., and Cass, G. R. (1993) Mathematical modeling of the concentrations of volatile organic compounds: Model performance using a lumped chemical mechanism, Environ. Sci. Technol. 27, Hildemann, L., Klinedinst, D. B., Klouda, G. A., Currie, L. A., and Cass, G. R. (1994) Sources of urban contemporary carbon aerosol, Environ. Sci. Technol., 28, Hoffmann, T., Odum, J. R., Bowman, F., Collins, D., Klockow, D., Flagan, R. C., and Seinfeld, J. H., (1997) Formation of organic aerosols from the oxidation of biogenic hydrocarbons, J. Atmos. Chem., 26, Izumi, K. and Fukuyama, T. (1990) Photochemical aerosol formation from aromatic hydrocarbons in the presence of NO x. Atmos. Environ., 24, Jacobson, M. Z. (1994) Developing, coupling, and applying a gas, aerosol, transport, and radiation model to study urban and regional air pollution, Ph.D. thesis, Department of Atmospheric Sciences, University of California, Los Angeles, California. Jacobson, M. Z. (1997) Development and application of a new air pollution modeling system. Part II: Aerosol module structure and design, Atmos. Environ., 31, Lurmann, F. W. and Kumar, N. (1996) Development of chemical transformation algorithms for annual PM10 dispersion models, draft report STI DFR, South Coast Air Quality Management District, Diamond Bar, California. Lurmann, F. W., Wexler, A. S., Pandis, S. N., Musarra, S., Kumar, N., and Seinfeld, J. H. (1997) Modeling urban and regional aerosols: II. Application to California s South Coast Air Basin, Atmos. Environ., 31, McMurray, P. H. (1989) Organic and elemental carbon size distributions of Los Angeles aerosols measured during SCAQS, Report to the Coordinating Research Council and the California Air Resources Board (CARB Contract A : 1989). McMurray, P. H., and Zhang, X. Q. (1989) Size distributions of ambient organic and elemental carbon, Aerosol Science and Technology, 10, Meng, Z., Dabdub, D., and Seinfeld, J. H. (1998) Size- and chemically-resolved model of atmospheric aerosol dynamics, J. Geophys. Res., 103, Moucheron, M. C. and Milford, J. (1996) Development and testing of a process model for secondary organic aerosols, Air & Waste Management Association Annual Meeting, Paper 96-FA , Nashville, Tennessee. Mylonas, D. T., Allen, D. T. Ehrman, S. H., and Pratsinis, S. E. (1991) The sources and size distributions of organonitrates in Los Angeles aerosol, Atmos. Environ., 5,

14 Odum, J. R., Hoffmann, T., Bowman, F., Collins, D., Flagan, R. C., and Seinfeld, J. H. (1996) Gas/particle partitioning and secondary organic aerosol yields. Environ. Sci. Technol., 30, Odum, J. R., Jungkamp, T. P. W., Griffin, R. J., Forstner, H. J. L., Flagan, R. C., and Seinfeld, J. H. (1997a), Aromatics, reformulated gasoline, and atmospheric organic aerosol formation, Environ. Sci. Technol., 31, (1997). Odum, J. R., Jungkamp, T. P. W., Griffin, R. J., Flagan, R. C., and Seinfeld, J. H. (1997b) The atmospheric aerosol-forming potential of whole gasoline vapor. Science, 276, Pandis, S. N., Paulson, S. E., Seinfeld, J. H., and Flagan, R. C. (1991) Aerosol formation in the photooxidation of isoprene and beta-pinene. Atmos. Environ., 25, Pandis, S. N., Harley, R. A., Cass, G. R., and Seinfeld, J. H. (1992) Secondary organic aerosol formation and transport. Atmos. Environ., 26, Pandis, S. N., Wexler, A. S., and Seinfeld, J. H. (1993) Secondary organic aerosol formation and transport II. Predicting the ambient secondary organic aerosol size distribution. Atmos. Environ., 27, Pankow, J. F. (1994a) An absorption model of gas/particle partitioning of organic compounds in the atmosphere. Atmos. Environ., 28, Pankow, J. F. (1994b) An absorption model of gas/aerosol partitioning involved in the formation of secondary organic aerosol. Atmos. Environ., 28, Pickle, T., Allen, D. T., and Pratsinis, S. E. (1990) The sources and size distributions of aliphatic and carbonyl carbon in Los Angeles aerosol, Atmos. Environ., 24A, Saxena, P. and Hildemann, L. M. (1996) Water-soluble organics in atmospheric particles: a critical review of the literature and application of thermodynamics to identify candidate compounds, J. Atmos. Chem., 57, Seigneur, C., Pai, P., Louis, J. F., Hopke, P., and Grosjean, D. (1998) Review of threedimensional air quality models for particulate matter, Report No. 4669, American Petroleum Institute, Washington, D.C. Seinfeld, J. H. and Pandis, S. N. (1998) Atmospheric Chemistry and Physcis, John Wiley, New York. 14

15 Stockwell, W. R., Middleton, P., Chang, J. S., and Tang, X. (1990) The second-generation regional acid deposition model chemical mechanism for regional air quality modeling, J. Geophys. Res., 95D, 16,343-16,367. Strader, R., Gurciullo, C., Pandis, S., Kumar, N., and Lurmann, F. W. (1998) Development of gas-phase chemistry, secondary organic aerosol, and aqueous-phase chemistry modules for PM modeling, Draft Final Report STI DFR, Coordinating Research Council, Atlanta, Georgia. Sun, Q. and Wexler, A. S. (1998) Modeling urban and regional aerosols near acid neutralityapplication to the June SCAQS Episodes, Atmos. Environ., submitted. Turpin, B. J., Huntzicker, J. J., Larson, S. M., and Cass, G. M. (1991) Los Angeles summer midday particular carbon: Primary and secondary aerosol, Environ. Sci. Technol., 25, Turpin, B. J. and Huntzicker, J. J. (1991) Secondary formation of organic aerosol in the Los Angeles basin: A descriptive analysis of organic and elemental carbon concentrations, Atmos. Environ., 25A, Turpin, B. J., Saxena, P., and Andrews, E. (1999) Measuring and simulating particulate organics in the atmosphere: Problems and prospects, Atmos. Environ., in press. Venkatram, A., Karamchandani, P., Pai, P., Sloane, C., Saxena, P., and Goldstein, R. (1997) The development of a model to examine source-receptor relationships for visibility on the Colorado Plateau, J. Air Waste Manage. Assoc., 47, Wang, S. C., Paulson, S. E., Grosjean, D., Flagan, R. C., and Seinfeld, J. H. (1991a) Aerosol formation and growth in atmospheric organic/no x systems I. Outdoor smog chamber studies of C 7 - and C 8 -hydrocarbons. Atmos. Environ., 26, Wang, S. C., Flagan, R. C., and Seinfeld, J. H. (1991b) Aerosol formation and growth in atmospheric organic/no x systems II. Aerosol dynamics. Atmos. Environ., 26, Yu, J., Flagan, R. C., and Seinfeld, J. H., Identification of products containing -COOH, -OH, and -C=O in atmospheric oxidation of hydrocarbons, Environ. Sci. Technol., 32, Zhang, S. H., Shaw, M., Seinfeld, J. H., and Flagan, R. C., Photochemical aerosol formation from α- and β-pinene, Geophy. Res., 97,

16 ATMOSPHERIC CHEMISTRY AND CHEMICAL MECHANISMS Draft as of 4/11/99 W.P.L. Carter 1, D.R. Crosley 2, D.M. Golden 2, L.T. Iraci 2, J.C. Johnston 2, and P.A. Makar 3 1 University of California, Riverside 2 SRI International 3 Environment Canada CONTENTS INTRODUCTION... 1 OVERVIEW OF ATMOSPHERIC VOC OXIDATION... 2 CURRENT STATE OF KNOWLEDGE... 4 Inorganic Reactions...4 Organic Reactions...5 Theoretical Estimates... 7 CHEMICAL MECHANISMS... 8 Summary of Chemical Mechanisms Currently in Use... 9 Lumping Techniques for Atmospheric Chemical Mechanisms Environmental Chamber Evaluations Mechanism Intercomparisons Variation in Model Predictions due to Photolysis Parameterizations and NMHC Reaction Mechanisms (Olson et al., 1997) Variations due to NMHC Chemistry (Kuhn et al., 1998) CURRENT STATUS FOR REACTIVITY MODELING CONCLUSIONS REFERENCES Note: Authors names are listed in alphabetical order INTRODUCTION In this chapter we summarize the status of understanding of the gas phase chemistry and photochemistry that is the basis of the mechanisms used in models of the chemical transformations involved in ozone formation. Given that gas phase processes lead to the formation of secondary aerosol particles and that verification of gas-phase mechanisms is often 1

17 based on smog chamber data that is influenced by wall effects, we discuss some heterogeneous chemistry in this section as well. OVERVIEW OF ATMOSPHERIC VOC OXIDATION The gas phase chemistry important in photochemical smog has been the subject of much study over the last fifty years. Our current understanding of the elementary reactions are given in various reviews and evaluations (Atkinson, 1989; 1990; 1991; 1994; 1997; Atkinson and Carter, 1984; Atkinson et al., 1997; NASA, 1997), the most recent being the NARSTO assessment of the atmospheric chemistry of VOCs and NO x prepared by Atkinson (1999). The discussion in those documents will be only briefly summarized here. The oxidation of hydrocarbons begins with the abstraction of a proton by the hydroxyl radical. In the presence of NO x, the subsequent reactions result in the conversion of molecular oxygen to ozone, as illustrated below for a general alkane. RH + OH R + H 2 O R + O 2 RO 2 RO 2 + NO RO + NO 2 2{NO 2 + hν NO + O} 2{O + O 2 O 3 } RO + O 2 Carbonyl Compound + HO 2 HO 2 + NO OH + NO 2 RH + 4O 2 +2hν H 2 O + 2O 3 + Carbonyl Ozone production continues as long as sufficient NO x is present so that reactions of peroxy radicals (RO 2 ) with NO compete effectively with their reactions with other peroxy radicals. Note that the OH radical levels are particularly important in affecting the O 3 formation rate in the presence of NO x because reaction with OH is a major (and in many cases the only) reaction pathway for VOCs. Thus, if a VOC reacts in such a way that it initiates radical levels (or forms a product that does), it would enhance the rate of ozone formation from all VOCs present. This would result in a larger effect on O 3 than other VOCs that react at the same rate. If the VOC s reactions in the presence of NO x have a radical termination process, it will cause all other VOCs to react more slowly and form less O 3. In some cases, this reduced O 3 formation from other VOCs may be more than enough to counter the ozone formation from the VOC s direct reactions. In such cases the VOC would have a negative effect on the formation of O 3 in the presence of NO x (Carter and Atkinson, 1989; Carter, 1994). Although an OH reaction is the major atmospheric loss process for most VOCs, some VOCs are also consumed to a nonnegligible extent by reaction with O 3 or NO 3 or by direct 2

18 photolysis. In most cases, these processes will also form RO 2 radicals, which convert NO to NO 2. In addition, and perhaps more significantly, many of these processes initiate the formation of new radicals, which ultimately cause higher OH radical levels and thus higher rates of reactions of the other VOCs present. This is particularly significant in the case of compounds that can photolyze, because photolysis reactions are the main sources of radicals in photochemical smog. For example, it is because of photolysis that formaldehyde has a much larger effect on ozone than one would estimate based on its OH rate constant alone. Ozone formation stops once NO x is consumed to sufficiently low levels. NO x is removed from the atmosphere more rapidly than total VOCs, since the NO x + OH rate constant exceeds that of most hydrocarbon + OH rate constants, and since the NO x removal processes generally involve a single step (such as the reaction of OH with NO 2 ) while most VOC reactions form products which are also reactive VOCs,. Therefore, NO x availability ultimately limits O 3 formation. If the NO x levels are high enough that it is not consumed before the end of the day, it is mainly the rate of the VOC s reactions, and their effects on OH radicals, which affect ozone levels. Indeed, high levels of NO x inhibit O 3 because reaction of OH with NO 2 reduces OH levels. High nighttime NO x levels may also reduce ozone, via conversion to NO 3 and HNO 3. If, however, NO x is consumed before the end of the day, O 3 is NO x -limited, and increasing NO x would cause increased O 3 formation. Under such conditions, if a VOC s reactions caused NO x to be removed more rapidly than if the VOC were absent (such as, for example, by forming nitrogen-containing products such as PANs from aldehydes and nitrophenols from aromatics), this would have a negative effect on O 3 yields, and tend to reduce the amount of O 3 formation caused by the VOCs reactions. Under highly NO x -limited scenarios, this becomes sufficiently important to cause VOCs with significant NO x sinks in their mechanisms to have negative effects on final O 3 yields even for those that may have highly positive effects on O 3 under conditions where NO x is plentiful. Another factor affecting the behavior of VOCs and NO x in ozone formation is competition for the hydroxyl radical. When the instantaneous VOC-to-NO 2 ratio is sufficiently low, OH reacts predominantly with NO 2, removing radicals and retarding ozone formation. Under these conditions, a decrease in NO x concentration favors ozone formation. At a sufficiently low concentration of NO x, or a sufficiently high VOC-to-NO 2 ratio, a further decrease in NO x favors peroxy-peroxy reactions, which retard ozone formation by removing free radicals from the system. Although, in general, higher VOC concentrations mean more ozone, increasing NO x may lead to either more or less ozone depending on the prevailing VOC-to-NO x ratio. As a result, the rate of ozone production is not simply proportional to the amount of NO x present; at a given level of VOC, there exists a NO x concentration at which a maximum amount of ozone is produced, or an optimum VOC-to-NO x ratio. Using an average VOC-OH reaction rate constant, representing reactions occurring in an average urban mix of VOCs, the ratio of the OH-NO x to OH-VOC rate constants is about 5.5. Thus, this optimum VOC-to-NO x ratio is approximately 5.5:1 for an average urban area, with the VOC concentration expressed on a carbon atom basis. For ratios less than this optimum ratio, NO x increases lead to ozone decreases, while at ratios higher than this optimum ratio, NO x increases lead to ozone increases. 3

19 Thus it can be seen that there are many mechanistic factors which must be appropriately represented in models used to predict the effects of a VOC on ozone formation. The specific mechanisms used to represent these processes in airshed models are discussed in a later section. First, we will give a brief summary of the current state of knowledge of the various types of reactions involved. CURRENT STATE OF KNOWLEDGE Inorganic Reactions In contrast to the many remaining uncertainties in our knowledge of VOC chemistry, reactions involving non-carbon-containing species are more thoroughly understood. The inorganic reactions are continuously reviewed by the NASA Stratospheric Data Panel (NASA, 1997) and the IUPAC Panel (Atkinson et al., 1997). Very recent data from the NOAA Aeronomy Laboratory (Ravishankara, 1998, personal communication) has elucidated the quantum yield for O( 1 D) formation from ozone in the 320 nm region. Also, new data (Brown et al.,1998; Dransfield et al., 1998) and re-evaluations of older data (Williams and Golden, 1998, personal communication) for the reaction of OH+NO 2 have removed much of the uncertainty associated with this reaction. However, current evaluations still give different recommendations concerning this extremely important reaction. One serious uncertainty in the knowledge of inorganic species concerns the formation of HONO. This important species, which photolyzes to produce OH, is probably formed heterogeneously in the atmosphere and in smog chambers. The characterization of this heterogeneous process is crucial, both for modeling the atmosphere and for understanding smog chamber data. Several studies have examined the heterogeneous formation of HONO for a variety of aerosol types. The most likely reaction pathway for HONO formation is via 2NO 2 + H 2 O HONO + HNO 3, rather than NO 2 + NO (N 2 O 3 ) + H 2 O 2 HONO (Andres-Hernandez et al., 1996; Kleffmann et al., 1997). The formation rate is highly dependent on the aerosol substrate and its age. Laboratory determined aerosol uptake coefficients for NO 2 range from 10-1 on freshly formed flame soot (Gerecke et al., 1997), to 10-4 on aged soot (Kamm et al., 1997) to 10-6 on aqueous aerosols (Kleffmann et al., 1997). The observed rapid change in soot aerosol HONO formation rates with time (Ammann et al., 1998) may indicate that chamber HONO formation may be strongly affected by the availability of reducing surface molecules on the chamber walls. Studies of HONO production on sulfuric acid and aqueous droplets (Iraci and Tolbert, 1997; Bambauer et al., 1994) suggest that heterogeneous processes occurring outside urban areas may also affect the HONO budget. The heterogeneous processes involving HONO formation are probably different in the atmosphere than in environmental chambers, but the extent to which they are different is unclear and represents an uncertainty in the use of chamberevaluated mechanisms in atmospheric simulations. Another important process, which is presumed to be primarily heterogeneous, is the hydrolysis of N 2 O 5 to HNO 3. This reaction can affect ozone yields because it can be a non- 4

20 negligible ozone sink, because it amounts to the formation of relatively unreactive HNO 3 from the reaction of ozone with NO 2. The gas-phase reaction is believed to be relatively slow, but the heterogeneous processes are non-negligible in chamber experiments, and their importance in the atmosphere varies with environmental conditions. Recent ambient atmosphere measurements and simulations (Makar et al., 1998c) indicate substantial losses of nitrate to the particle phase following the reaction of NO x with O 3 at night. The formation rate of gas-phase HNO 3 was significantly enhanced through the effects of heterogeneous chemistry. Organic Reactions The most complex and uncertain area of atmospheric chemistry is the photooxidation reactions of the many types of VOCs that can be emitted. Our current state of knowledge of the atmospheric reactions of VOCs is discussed in various reviews by Atkinson (1989; 1990; 1991; 1994), the most recent being the NARSTO assessment (Atkinson, 1999). Current European laboratory work can be found in the proceedings of the Eurotrac-2 Chemical Mechanism Development Subproject meetings (Ammann and Lorenzen, 1997). The aspects of the organic reactions that must be considered are the rate constants, the photooxidation steps, and the mechanisms of the products formed. These are summarized briefly below. The least uncertain aspect of the atmospheric reactions of the organic compounds concerns their initial rates of reaction with OH or NO 3 radicals or with ozone. Many references appear in the Atkinson reviews (Atkinson, 1989; 1991; 1994; Atkinson and Carter, 1984) and in the NASA and IUPAC compilations (Atkinson et al., 1997; NASA, 1997). Modern rate constant measurements are often precise, and where individual rate constants have been measured they are often known fairly well. Nevertheless, the stated uncertainties in rate constants in the compilations are almost always 25% or greater and higher uncertainties should be assumed if there is only a single measurement, and if the compound is of low volatility or has high surface affinity. For most compounds it is usually not particularly difficult or costly to obtain these rate constants if no data are available. Methods exist for estimating rate constants for the reactions of VOCs with OH and NO 3 radicals which can be used when data are not available (e.g., Kwok and Atkinson, 1995; Atkinson, 1997). These estimates may be good to +/- 50% in the most favorable cases, but a factor of 2 is probably a more realistic uncertainty estimate for most VOCs. Obviously the estimates are probably not reliable if the VOC has functional groups which have not been studied. Also, it should be noted that ozone rate constants appear to be difficult to estimate reliably (see Atkinson and Carter, 1984). Data concerning rate constants for the reactions of the radical intermediates are much more limited and are usually restricted to the simplest cases. It has been assumed that the higher molecular weight radicals react with the same rate constant. 5

21 An interesting question is the extent to which reactions that are pressure dependent, such as radical-radical processes, have been treated correctly in the chemical mechanisms. In addition, the special nature of H 2 O as a third body may need to be taken into account. A very important process involving the reaction of peroxy radicals with NO remains to be understood on a fundamental level. This interaction seems to follow two competing pathways. One pathway forms alkoxy radicals and NO 2, perpetuating the free radical chemistry, whereas the other forms organic nitrates and removes radicals from the system. The branching ratio for this process has a significant effect on a VOC s ozone reactivity because organic nitrate formation is a radical and NO x sink process, while the alkoxy + NO 2 branch is neither. This branching ratio has been measured accurately for only a few types of peroxy radicals; for most VOCs this has to be treated as an adjustable parameter, or the existing product yield data must be extrapolated to similar compounds. Much of the complexity in organic photooxidation mechanisms comes from the various alternative reactions alkoxy radicals can undergo. These include, but may not be limited to: (1) reactions with O 2 forming HO 2 and the corresponding carbonyl compound (for radicals with alpha hydrogens), (2) beta-scission decomposition forming a carbonyl compound and another radical; and (3) hydrogen shift isomerizations forming a hydroxy substituted radical. More recently, Tuazon et al. (1998) found that alkoxy radicals formed in the photooxidations of esters can undergo a previously-unsuspected ester rearrangement reaction involving hydrogen transfer to the carbonyl group. Absolute rate constants have been measured for only a few of the most simple alkoxy radical reactions, and most of the other available data concerns ratios of rate constants which can be inferred from results of product studies. This type of information is becoming available for an increasing number of systems because of ongoing product studies, though these branching ratios still need to be estimated for the large majority of VOCs emitted into the atmosphere. Based on the limited information available, Atkinson (1997) developed methods for estimating rate constants (or rate constant ratios) for the alkoxy and peroxy radical reactions occurring in the alkane and alkene + OH reaction systems. This was recently extended by Carter (unpublished work, 1998) to the reactions of other classes of compounds, primarily various oxygenates (this is discussed further below). Although these estimates provide means to derive mechanisms which represent the best estimate given available data, the estimation methods only approximately fit the available data. For example, it would not be unexpected for the nitrate yield estimates for RO 2 +NO reactions to be off by 50% to a factor of 2, or for the alkoxy radical branching ratios to be off by a factor of 5. As discussed by Atkinson (1999), much greater uncertainties are involved in our understanding of the atmospheric reaction mechanisms of aromatic hydrocarbons and the mechanisms for the reactions of ozone with double bonds. Progress has been made in these areas, but the available information is far from sufficient to derive predictive chemical mechanisms for modeling ozone and other impacts of VOCs. In both cases, uncertain parts of the mechanisms have to be parameterized or adjusted so that model predictions are consistent with environmental chamber data. 6

22 Very limited information is available concerning the atmospheric reactions of compounds containing atoms other than C, H, or O. Although many halogenated compounds have been studied in the context of their impacts on remote atmospheres or the stratosphere, data on their ozone impacts are extremely limited, and recent studies of trichloroethylene and alkyl bromides show that current estimated mechanisms cannot successfully predict their reactivities in environmental chamber experiments. Ozone impacts of volatile siloxanes have also been studied, but the reactivity data obtained are difficult to reconcile with results of product studies (Hobson et al., 1997). The various areas where research is most needed concerning the atmospheric reactions of organics are summarized below. Most of these are from the conclusions of the NARSTO evaluation (Atkinson, 1999). Rate constants and mechanisms for reactions of peroxy radicals with NO, HO 2, other RO 2, and NO 2 radicals. This would include additional data for nitrate yields from peroxy + NO reactions, particularly for non-hydrocarbon reactions. Branching ratios for the competing reactions of alkoxy radicals, particularly those not formed from alkanes and alkenes. Details of the reactions of ozone with alkenes and other VOCs containing double bonds under atmospheric conditions. Total radical yields are particularly important in model simulations of VOC reactivity. Thermal decompositions and other atmospherically important reactions of the higher PAN analogues, such as that formed from methacrolein and isoprene. Mechanisms and products of the reactions of OH - aromatic adducts with O 2 and NO 3. Quantitative yield information and studies of the reactions (including photolysis) of these aromatic products are especially needed. Tropospheric chemistry of the oxygenated products formed from the radical NO x and radical radical reactions in the photooxidation of the VOCs requires study. Quantitative understanding of reaction sequences leading to secondary organic aerosol formation. Information concerning reactions of halogen-containing radicals under tropospheric conditions is also needed before reactivities of halogen-containing compounds can be assessed with any accuracy. Information concerning the reactions of radicals formed from the reactions of amines and other nitrogen-containing compounds is needed before reactivities of such compounds can be assessed with any accuracy. Theoretical Estimates In recent years the development of ab initio theoretical methods for the calculation of potential energy surfaces allows the direct computation of some rate constants. (See for example Irikura and Frurip, 1998). Transition state theory can also be utilized in this regard. These 7

23 computational techniques have not yet been exploited to any significant extent in the uncertain areas of atmospheric chemistry, and the time seems right for a serious theoretical look at many of these processes. Some exploratory studies have used these methods to suggest mechanistic pathways for the photooxidation of napthalene (Lane et al. 1996). These theoretical techniques should be tested through comparison to known processes (e.g., oxidation of the lower C number alkanes and alkenes), then applied towards predicting mechanisms and reaction rates which are currently unknown. Ab initio methods may provide a useful means for reducing the time required for laboratory confirmation of these processes, by suggesting specific product compounds for analysis in chamber experiments. Estimates of heats of reaction are also used in many of the estimation methods referenced above, and often can be used to rule out chemically unreasonable reaction schemes. Group additivity methods based on the work of Benson (1976) are obviously very useful in this regard, but there are all too often groups for which no data are available. Theoretical calculations could potentially be very useful in providing the data needed to support application of thermochemically-based estimation methods and evaluating proposed reaction sequences. CHEMICAL MECHANISMS The chemical mechanism is the portion of the airshed models that represents the gasphase reactions discussed in the previous section. Because of the large number of compounds emitted or formed in polluted troposphere and the large numbers of reactions they, and their reactive products, can undergo, these mechanisms must necessarily contain significant simplifications and approximations. Furthermore, because of limitations in our knowledge, these mechanisms must necessarily contain assumptions and extrapolations to represent processes that are important but whose details are unknown. Different mechanism developers can apply different approaches to simplify the mechanism to make it tractable and can use different assumptions and extrapolation methods when representing the main areas of uncertainty. In the past the main limiting factor has been computer-related limitations, but this is becoming much less of a factor now as computers become more powerful and as software used to implement mechanisms become more capable and flexible. The main limitation now is our level of knowledge of the many processes which must be represented, and our ability to generate and manage highly complex reaction schemes in a manner that is appropriate given our level of knowledge. Obviously, it is not appropriate or a good use of computer power to use highly complex and explicit reaction schemes if the added complexity is speculation and the resulting predictions no more accurate than predictions from the highly condensed mechanisms used in the current generation of models. Given the development of the Morphecule approach of Jeffries and co-workers (see Dodge, 1999) and the mechanism generation approach being developed by Carter (summarized below), in the very near future our level of knowledge is going to be the main factor limiting the level of detail and size of the current mechanisms. Most, though not all, of the mechanisms used in the current generation of models have been summarized by Bergin et al. (1997) and Dodge (1999). The major mechanisms relevant to 8

24 current reactivity assessments are summarized below. When considering mechanisms to be used in reactivity assessment, the following issues need to be addressed: How is uncertainty dealt with? What are the approximations and lumping approaches used? How up-to-date are the rate constants used? To what extent have the mechanisms been evaluated? Summary of Chemical Mechanisms Currently in Use The Carbon Bond IV (CB4) mechanism (Gery et al, 1988, 1989) is important because it is widely used in regulatory models. Its rate constants and reaction schemes represent the state of knowledge as of approximately 1997, although some important rate constants have been updated since then (see Dodge, 1999). This uses a highly condensed method to represent reactions of individual VOCs, with the goal being to predict ozone from ambient mixtures as accurately as possible but with high computational efficiency. (Lumping techniques are discussed in more detail below.) It was evaluated against a large number of environmental chamber experiments (Gery et al, 1988), and was reasonably successful in predicting ozone formation from complex mixtures. However, it is not suitable for assessing reactivities of individual VOCs because of its high level of condensation, and some of the simplifications and approximations it employs are now believed to be inappropriate. The RADM-2 mechanism developed by Stockwell et al. (1990) is used in the EPA s Regional Acid Deposition (RADM) model and is the only mechanism currently incorporated in the EPA s Models-3 system. Its rate constants represent the state of knowledge as of It is a condensed mechanism which represents reactions of groups of similarly reacting VOCs with a single model compound with fixed parameters, so it, like CB4, is not strictly suitable for reactivity assessment of individual VOCs. It accounts for reactivity differences among individual VOCs in a given class by using reactivity weighting, where the amount of model compound used to represent the VOC is greater if the rate of reaction of the compound is greater, and vise-versa (see discussion of compression methods, below). A relatively limited number of model compounds are used to represent reactions of higher molecular weight organic oxidation products. However, it represents more classes of compounds using reactivity compression, which probably introduces fewer errors than the condensation approach used in CB4. This mechanism has the most detailed (and probably most accurate) representation of the low-no x peroxy radical reactions than most of the other mechanisms, including the SAPRC mechanisms discussed below. This mechanism was extensively evaluated against available chamber data by Carter and Lurmann (1990), and performed reasonably well in simulating ozone in experiments with complex mixtures and individual compounds which this mechanism is designed to represent. Its treatment of many of the more important uncertain reactions is similar to that of the SAPRC-90 mechanism, discussed below. The RADM-2 mechanism was recently updated and expanded by Stockwell et al. (1997) who renamed it the RACM ( Regional Atmospheric Chemistry Mechanism ). It is the most updated of the published mechanisms in terms of its rate constants and the mechanisms for its 9

25 explicit reactions. It has a similar condensation approach as RADM-2, though the number of classes of compounds which are represented separately have been increased. It was evaluated against a limited number of chamber experiments, and against the predictions of the more thoroughly evaluated RADM-2 mechanism. The SAPRC mechanisms are important because they are designed specifically for VOC reactivity assessment, and have been employed to generate reactivity scales which have been or are being considered for use in regulatory applications (Carter, 1994; CARB, 1993). Condensed versions of this mechanism have been adapted for use in Eulerian airshed models (e.g., Lurmann et al, 1991; see also references in the Reactivity Assessments section), but its primary use in reactivity assessments has been to calculate incremental reactivities in EKMA-type model scenarios (Carter, 1994). This mechanism can separately represent the reactions of over 100 different types of VOCs by using generalized reactions with variable parameters which are assigned based on the known or estimated rate constants and products of the compounds. This feature makes it particularly useful for assessing reactivities of a large number of VOCs. Nevertheless it uses a very condensed representation of the reactive organic oxidation products (though not as condensed as CB4, RADM-2 or RACM), and uses a much more condensed representation of peroxy + peroxy reactions than does RADM-2 or RACM (though not as condensed as CB4). The best documented version is SAPRC-90 (Carter, 1990), whose rate constants represent the state of the art as of 1989, and is thus approximately contemporary with CB4 and RADM-2. The SAPRC-90 mechanism has been evaluated against results of approximately 500 chamber experiments and in most cases fits the ozone data to within ±30%, which, in the case of complex mixtures representative of atmospheric conditions, is comparable to the performance of RADM-2 (Carter and Lurmann, 1991) or CB4 (Carter, unpublished results). However, the performance is obviously better in simulations of the individual compounds that SAPRC-90 represents explicitly, but which are not well represented by the model species used in the condensed mechanisms. The SAPRC mechanism has been updated several times since SAPRC-90, based on results of chamber experiments on individual compounds (e.g., Carter et al., 1993; Carter, 1995; Carter et al., 1997), though the updates have not been comprehensively documented in peerreviewed journals. The latest version incorporates a complete update of the rate constants and updated estimates for a variety of compounds, and slight improvements in the level of detail in representing reactive products and low-no x peroxy radical reactions. The automated procedure for generated alkane reaction mechanisms was updated based on the results of the evaluation of Atkinson (1997) and an independent evaluation of alkoxy radial reactions (Carter, unpublished results, 1998), and it was extended to include alkenes (with no more than one double bond), and many classes of oxygenates including alcohols, ethers, glycols, esters, aldehydes, ketones, glycol ethers. Although many of the estimated rate constants and rate constant ratios are highly uncertain (see discussion of atmospheric chemistry, above), this procedure provides a consistent basis for deriving best estimate mechanisms for chemical systems which are too complex to be examined in detail in a reasonable amount of time. The mechanism generation program allows for assigning or adjusting rate constants or branching ratios in cases where data are available, or where adjustments are necessary for model simulations to fit chamber data. Various lumping rules are used to convert the detailed generated mechanisms and product distributions into the 10

26 lumped reactions and model species distributions actually used in the model. The use of this program has permitted estimation of detailed mechanisms for a much larger number of compounds than otherwise would be possible. The latest updated SAPRC mechanism was evaluated using the indoor environmental data base, including relevant runs carried out very recently for VOC reactivity assessment (see reports at Uncertainty classifications were derived for the various classes of VOCs represented in the mechanisms. Additional information concerning this mechanism, the listing and uncertainty classifications of the VOCs which it can represent, and updated MIR and other reactivity calculations are available at An alternative to the SAPRC mechanisms for reactivity assessment are being developed and applied by researchers in Europe (e.g., Derwent and Jenkin, 1991; Andersson-Skold et al., 1992; Derwent et al., 1996; Jenkin et al., 1997). These mechanisms are based on the concept of representing the reactions, the significant or representative VOCs, and also their major or representative oxidation products, explicitly. Probably the most detailed of these is the Master Chemical Mechanism (MCM) of Derwent and co-workers, which can be seen at Although these mechanisms are nominally explicit, condensation is employed by excluding minor processes and products, in effect representing them by major or representative species. Also, as with all other mechanisms, no attempt is made to represent the unknown aromatic ring fragmentation products in detail. These mechanisms are not used in the United States because the model software is not adapted to mechanisms of this size. Also, they have not yet (to our knowledge) been evaluated against results of environmental chamber experiments. In Canada, the ADOM-II mechanism (Stockwell and Lurmann, 1989; Lurmann et al., 1986) is currently used for reactivity simulations. The mechanism went through several stages of development, from its initial creation as an urban-scale ozone prediction mechanism (Lurmann et al., 1986), updating and comparison to 490 chamber experiments (Carter et al., 1986; Lurmann et al., 1987), and updating the most condensed mechanism of Lurmann et al.(1987) to be consistent with Atkinson (1988) and Carter (1988) and adding species and reactions important for long-range transport and acid-deposition modeling (Stockwell and Lurmann, 1989). Further smog chamber testing of the ADOM-II mechanism is currently underway for comparison with the AURAMS mechanism, described below. A new Canadian mechanism is under development for gas-phase and particulate modeling in the AURAMS model (Moran et al., 1998) which will have updated rate constants and a greater level of chemical detail than the ADOM-II mechanism (Makar et al., 1998a). Some of the features of this mechanism currently include: A revised and detailed biogenic hydrocarbon mechanism, based on the detailed isoprene mechanism of Carter and Atkinson (1996), the photolysis data of Raber and Moortgat (1996) and Gierczak et al. (1997), and the alpha-pinene mechanistic data of Hakola et al. (1994) and Hatakeyama et al. (1989, 1991). Methacrolein and methylvinylketone are included, as are their oxidation pathways by OH, O 3 and NO 3, as is the formation of MPAN. Explicit RO 2 -RO 2 reactions between the generic isoprene organic radical, the 11

27 alpha-pinene organic radical, and the eighteen other organic radicals of the mechanism are included. In the context of biogenic chemistry, the mechanism was recently used to simulate the emission, transport and chemical loss of biogenic compounds with good agreement to ambient measurements (see Makar et al., 1998b for further mechanism description and references). An updated aromatic mechanism, with a simplified broken-ring oxidation pathway based on the work of Becker (1994), with muconaldehyde, oxybutanal and methylglyoxal as dicarbonyls formed from the broken-ring pathway. The non-ring-breaking pathway products include benzaldehyde and a generic aromatic nitric anhydride cycle. Toluene, di- and tri-methyl benzene are the three aromatic species resolved. Internal and terminal bond alkenes resolved as separate species, with separate radicals and pathways for NO 3 oxidation as well as OH oxidation. The biradical stabilization pathway for the alkenes (including the biogenic species isoprene and alpha-pinene) is based on the work of Horie et al. (1994). Five alkane species; methane, ethane, propane, C 4 -C 5 alkanes and C 6 -C 7 alkanes; the last being generic species. Alcohols up to C 3 are resolved, organic acids are represented by formic and acetic acid, as is peroxypropylnitrate. A single generic organic peroxide is resolved, although current plans are to include an additional generic hydroxy-organic peroxide to better simulate biradical stabilization following ozone oxidation of alkenes. Detailed, speciated organic radical reactions (RO 2 + RO 2, RO 2 + R(O)O 2, RO 2 + HO 2, RO 2 + NO 3, RO 2 + NO). The self-reactions and HO 2 reactions are of particular concern for low NO x environments with high VOC emissions, such as the boreal forests of Canada. The mechanisms and rates of these reactions are highly uncertain, and are based on extrapolation from the limited available laboratory data as well as the rate estimation procedures of Atkinson (1997b). Both the AURAMS mechanism and the ADOM-II mechanism have been compared to a limited number of smog chamber data (UNC chamber single species tests); further testing against complex mixtures and SAPRC data was to take place by March of The tests to date have shown that the new mechanism shows a significant improvement in the ability to predict chamber data compared to the ADOM-II mechanism. Further testing will take place using the SMVGEAR code (Jacobson and Turco, 1994) to facilitate rapid comparison to a large number of chamber runs. Lumping Techniques for Atmospheric Chemical Mechanisms The gas-phase reaction mechanisms used in predicting atmospheric reactivity frequently have simplified or compressed VOC kinetics. Detailed, highly speciated mechanisms are available, but their use can be impractical for large numbers of simulations, for either multiple box model calculations or the chemical integrations for a regional reaction/transport model. For some types of integration routines, the processing time required to perform a single chemical integration may be dependent on a power law function of the number of variables. The storage 12

28 of the mixing ratios of the hundreds to thousands of species found in the real atmosphere may also place a burden on the available computational resources. These combined limitations of processing time and memory space have resulted in the creation of simplified mechanisms for atmospheric chemistry. The main use for these simplified mechanisms has been the prediction of acid precipitation and ozone production. As a result, the mechanisms have attempted to preserve the reactivity of the simulated atmosphere, while using less model species than occur in the real atmosphere. One common aspect to all of the reduced mechanisms is the use of a smaller number of oxygenated species than is present in the real atmosphere. For example, the OH radical oxidation of a long chain alkane may create several organic radicals, in turn leading to the formation of several different carbonyls after reaction with NO. A compressed reaction mechanism may represent these species with a single organic radical and a single carbonyl. The rationale for this form of simplification is two-fold. First, the rate of the RO 2 + NO reaction is relatively invariant across different RO 2 species, hence a single RO 2 may be sufficient to accurately convert NO to NO 2 within the model. Second, the carbonyl species are assumed to have a secondary importance to the initial hydrocarbon with regards to ozone formation, and that simplifications to broad classes of oxygenated species are therefore justifiable. The second assumption is weaker than the first; recent laboratory work on the mechanistic pathways of species like the aromatic compounds have shown that the oxygenated product species can be very reactive, with the reactivity varying widely for the different carbonyls formed. This has resulted in increased speciation of oxygenated compounds in mechanisms which have been recently published (e.g., RACM; Stockwell et al., 1997) or are under development (Canadian AURAMS mechanism), compared to their predecessors. The unoxygenated species have been reduced (or lumped or compressed ) using several methods, usually in two to three stages. The detailed, speciated emissions are combined into a smaller number of species representing broad chemical classes, which are then combined to the model speciation using reactivity weighting (cf., Middleton et al., 1990, and the Emissions section of this report). Finally, in the chemical mechanism itself, some form of lumping is used to attempt to create the same product distribution as in the unlumped mechanism. This stage deals with the question of how to combine reactions such as {hydrocarbon 1} + OH a {product 1}, k 1 {hydrocarbon 2} + OH b {product 2}, k 2 to give a net reaction: {lumped hydrocarbon} + OH A {product 1} + B {product 2}, k 3 The focus of the problem being how to determine a net reaction rate constant k 3 and new product coefficients A and B which have the same effect on the OH, net hydrocarbon, and product mixing ratios as the original system of two (or more) reactions. 13

29 Different methods for mechanism compression include those based on reactivity, concentration weighting, and reactivity across carbon bonds within the molecules of each individual species. Bond (CB4) and concentration-based compression methods are described in (Dodge, 1999). Several papers have been published on the mathematical aspects of reactivitybased compression. These compare the lumping methods used in the ADOM-II mechanism (Stockwell and Lurmann, 1989; Lurmann et al., 1986), and the RADM2 mechanism (Stockwell et al., 1990), and devising more accurate reactivity-based methods (cf., Makar et al., 1996; Makar and Polavarapu, 1997; Makar, 1998). The earlier reactivity-based methods made use of an average OH concentration integrated over time, and knowledge of the hydrocarbon solution at either small or infinite times to form approximate product yields (Makar et al., 1996). An improved approximate solution using both small and long time scale limits was proposed (Makar et al., 1996), but subsequent work (Makar and Polavarapu, 1997; Makar, 1998) has shown that an arbitrarily large number of unoxygenated hydrocarbons with more than one oxidant can be compressed with no loss in accuracy. In addition, the earlier techniques such as that used in the RADM mechanism could sometimes lead to large underpredictions in the ozone mixing ratio. The technique has recently been expanded to oxygenated species (Makar, 1998). One potential use of the compression numerics is to compress the mechanism in a transient fashion; retain the detailed speciation until chemical integration is required, then compress the mechanism for the purposes of integration. Post-integration, the original information may be recovered. This concept of temporary mechanism compression has appeared twice in the literature, in the context of lumping by reactivity (Makar, 1998) and by concentration (Morphecule mechanism; Dodge, 1999). A comparison between these methods might be worthwhile, given their similar aims yet different mathematical approaches (Concentration weighting may lead to errors, depending on the relative reactivities of the compressed species, and the case of multiple oxidants needs to be considered. The details of the morphecule mathematics were not given in, 1998, precluding a comparison here). The advent of these methods will allow increased hydrocarbon speciation in future modeling of reactivity. At the same time, as noted in Dodge (1999), increased model speciation, created in the absence of laboratory based mechanistic or kinetic data, will add little confidence to model results. In addition, the ozone forecasts from mechanisms with a variety of complexities (from highly parameterized to very detailed) has been shown to have a relatively minor effect on the magnitude of ozone produced (Kuhn et al., 1998; see also the section on model intercomparisons of chemical mechanisms). The extent to which the use of detailed, temporarily compressed mechanisms improves ozone simulations has yet to be determined, and would be another area worthwhile of further study. As indicated above, most current mechanisms use a limited number of model species to represent the large number of higher molecular weight oxidized product species. An indication of the importance of this was obtained during the latest update of the SAPRC mechanism, when a new model species was added to represent the reactions of the more reactive ketones and other non-aldehyde oxygenated products (previously MEK was used for all these products.) It was found that this modification caused an increase of approximately 30% in the calculated MIR for compounds such as glycol ethers, even though it caused almost no change in the model simulations of the incremental reactivity chamber experiments with those compounds (Carter, unpublished results). This indicates the importance of accurate representation of oxidation products. 14

30 Environmental Chamber Evaluations Before any chemical mechanism whether detailed or condensed is incorporated in an airshed model, it must be demonstrated to predict at least the major features of the VOC-NOx-air photooxidation process. The only practical means for doing this is to conduct experiments using an environmental chamber, also called a smog chamber, where the chemical processes of interest are occurring under controlled and well-characterized conditions. It can then be determined whether the experimental results are consistent with the predictions of the chemical mechanism. Chemical mechanism development experiments have been performed in indoor chambers of approximately liters using artificial light sources (Carter et al., 1995), much larger outdoor chambers (Jeffries et al., 1982; Wang et al., 1992; Odum et al., 1996), and with smaller indoor reaction bags (Kelly et al., 1994; Kelly and Wang, 1996). Various types of chamber experiments are used to test different aspects of the chemical mechanisms. Irradiations of single VOCs in the presence of NO x and air test the mechanisms for the individual compounds; NO x -air irradiations of more complex VOC mixtures test the performance of the model as a whole and experiments where the effect of adding single VOCs to irradiations of NO x and complex mixtures test model predictions of the VOC s incremental reactivity. Evaluation of chemical mechanisms with chamber data is complicated by uncertainties in chamber effects (Carter and Lurmann, 1990; 1991; Jeffries et al., 1992), and separate characterization experiments are needed to evaluate models for these effects. Although this introduces uncertainties in such evaluations, the uncertainties in evaluating chemical mechanisms using chamber data are far less than the uncertainties in attempting to evaluate mechanisms by comparing full airshed modeling results with ambient air data. With chamber experiments, the amounts of input pollutants are accurately known, and no uncertainties regarding dilution or transport need to be considered. Current chamber data are available to test the mechanisms for only a subset of the many types of VOCs emitted into the atmosphere. However, ongoing studies, motivated largely by the need for data to support potential VOC exemption petitions or the need to reduce uncertainties so reactivity considerations in VOC regulations, is resulting in an ever-increasing number of compounds for which environmental chamber data are available for mechanism evaluation. Although there are really no practical alternatives at the present time, use of environmental chambers for mechanism evaluation is not without significant problems. Given that smog chamber experiments are carried out under conditions where trace species are much more concentrated than under ambient conditions, great care must be taken in extrapolating the smog chamber data to atmospheric conditions. Perhaps the time derivative of the ozone concentration would be a more appropriate indicator than absolute ozone concentration. An important point to consider is whether or not the chemical mechanism is reproducing the correct ozone behavior for the right reasons. The comparison of additional species would allow for more thorough evaluation of the mechanism and the possibility of counterbalancing errors. Another issue worth addressing is whether total oxidants (O 3 + H 2 O 2 + ROOH) would be a more complete way of diagnosing the reactivity of a particular VOC than just looking at ozone formation or ozone formation + NO oxidation, as is usually employed. If downstream effects are 15

31 important, the first round of organic products needs to be predicted and tracked accurately so their impact during and after multi-day transport can be assessed. Important analytical issues and some possible shortcomings of chamber experiments are listed below: Intensity and spectral characteristics of the chamber light sources are difficult to characterize. It is difficult to perform experiments at low concentrations, so experiments are usually not directly representative of environmental conditions. Additionally, the VOC/NO x ratios are usually higher than in the atmosphere. Wall reactions that can be the principle source of radicals are not understood. How well are temperature and relative humidity monitored in the smog chambers? Are these parameters included in the models? Are fast analytical techniques available for monitoring appropriate intermediate species? Carbonyls, if measured, are often collected using cartridge techniques. What is the integration time for the PAN GC/ECD technique? There are problems with complex mixtures where components cannot be completely identified. Low volatility may lead to decreased concentration of a VOC in the gas phase, but will heterogeneous reactions on aerosols or in the soil release product species that may lead to smog formation? Few multi-day chamber experiments are available for testing downstream predictions. Chamber experiments tend to be much less sensitive to the representation of the reactions of reactive organic products than do model simulations of the atmosphere. This is because the integrated OH radical levels tend to be lower in current chamber experiments than in atmospheric scenarios. Mechanism Intercomparisons The comparison of the predictions of different tropospheric reaction mechanisms for a common set of initial and boundary conditions is a useful means of identifying factors affecting the accuracy of reactivity simulations. The intercomparisons also highlight sources of reactivity simulation uncertainty and needs for further laboratory work. Two recent papers (Kuhn et al., 1998; Olson et al., 1997) have examined several mechanisms in this fashion; their results are summarized here. Variation in Model Predictions due to Photolysis Parameterizations and NMHC Reaction Mechanisms (Olson et al., 1997) The reaction rates for non-methane hydrocarbon (NMHC) chemistry from twenty-one modeling groups were compared for a common set of atmospheric and radiative parameters and 16

32 initial conditions (clear sky, solar zenith angle = 23 o, latitude = 45 o N, July 1, US Standard Atmosphere, 4 NMHC initial concentration regimes). Sixteen of the groups also examined NMHC effects in two addition test cases. The diurnal averages of the photolysis rates for the reactions O 3 + hν O( 1 D) + O 2, H 2 O 2 + hν 2 OH, NO 2 + hν NO + O( 3 P) and HCHO + hν H + CHO were calculated by the different models and compared at four altitudes. Ozone and hydrogen peroxide photodissociation mean values were essentially identical (with overlapping rms errors). NO 2 and formaldehyde photolysis rates for models employing multistream methods were significantly larger than two-stream models. Multistream NO 2 photodissociation mean values were about 10 s -1 larger than two stream rates at all levels (an increase of approximately 20%), while formaldehyde rates increased by approximately 25%. Five day diurnal box model simulations were used to compare the mixing ratios of O 3, NO, H 2 O 2 and diurnal values of HO 2 and OH. For simulations lacking NMHCs, variations in O 3 and NO x predicted by the different models was small; less than 5%. With in the inclusion of NMHC s O 3 and NO x means and medians diverged by up to 25% Analysis of the surface level no-nmhc cases showed that the models fell into three subsets based on HO x and O 3 mixing ratios. Models lacking the pressure and water-vapor pathways of the HO 2 self-reactions tended to have lower H 2 O 2 mixing ratios in favor of increased HO 2. Differences in HO x between the remaining two subsets were attributed to differences in the H 2 O 2 and HCHO reservoir concentrations, in turn dependent in part on photodissociation rates. The case of a NO x (no VOC) plume had O 3 differences of 11% after 5 days of simulation, attributed to differences in the ozone photolysis rate predicted by the different models. The hydrocarbon test runs showed a wider variation in model results, due to the different hydrocarbon oxidation schemes used. The hydrocarbon schemes could be grouped into three categories, depending on their original source and form of chemical lumping. These categories included: those mechanisms based on the lumped molecule approach (e.g. RADM-II; Stockwell et al., 1990), the lumped molecule with surrogate species approach (e.g. Lurmann et al, 1986), and lumped structure mechanisms (e.g., Carbon-Bond IV; Gery et al., 1989). The choice of a lumping method had no consistent effect on the predicted O 3 or NO x concentrations; differences in predictions did not correlate with the mean values of compression used in the given mechanism. The NMHC tests showed a much larger rms error for NO x between mechanisms (40% vs. 15%) than the no-nmhc tests, and the rms error for O 3 doubled. Although much of the model variation was attributed to differences in the photolysis calculations, the authors noted that the 5-15% rms variation about the mean for the photodissociation rates was within the range of accuracy of the measurements of quantum yields and cross-sections (cf., DeMore et al., 1992). 17

33 Variations due to NMHC Chemistry (Kuhn et al., 1998) Twelve different mechanisms were compared in several tests, including one (PLUME1) in which photolysis rates were prescribed and typical emission levels for continental European air were included as first order production terms. The mechanisms included four variations on the RADM2 mechanism, three derived from the Carbon Bond-IV mechanism, the EMEP mechanism, the ADOM-II mechanism, and four explicit schemes. As in the other intercomparison noted above, the pressure and water vapor pathways of the HO 2 self reaction were not incorporated in all mechanisms, sometimes leading to significant differences in H 2 O 2 predictions. Other causes included variations in the rate constants for organic peroxides with OH, and the level of detail with which RO 2 and HO 2 reactions were treated in each mechanism. The rms variation between different mechanisms for O 3 production in the PLUME1 case was 16% of the mean, with individual mechanisms being up to 27% higher than the mean (EMEP) and 35% less than the mean (CB4-TNO) (See Figure 1, below). The authors noted that the NO x emission level had a greater effect on O 3 than VOCs for the simulations, with the need for improved treatment of HO 2 RO 2 interactions and peroxide formation having a significant effect on model results. This was identified as a weak point in many reaction mechanisms. Other sources of model differences included the extent of speciation (e.g., the 1640 reaction, 715 species IVL scheme had larger higher aldehyde concentrations than the other schemes due to increased speciation). Mixing ratios of H 2 O 2, organic peroxides, and higher aldehydes all had higher rms errors, with deviations about the mean of 30%, 56%, and 50%. Formaldehyde values Figure 1 (From Kuhn et al, 1998) Left: Ozone concentrations over 5 days of diurnal chemistry with emissions, results from 12 mechanisms. Right: Formaldehyde concentrations, results from 12 mechanisms. 18

34 (Figure 1) varied by up to +63% to -67% of the mean value by the end of the five day diurnal simulation. One of the main conclusions from (Kuhn et al., 1998) was that similar O 3 mixing ratios were predicted by the different mechanisms, despite differences in the hydrocarbon parameterization. The use of more detailed oxygenated hydrocarbon mechanisms tended to result in higher ozone predictions, due the increase in the RO 2 production associated with these species. OH radical differences were also small (rms difference of 19% for the Plume 1 case). Concentrations of the longer lived species varied considerably. As noted by Kuhn et al., (1998), model intercomparisons are not sufficient for determining the reliability of a mechanism in predicting actual ozone mixing ratios; this must be done through comparisons to measurements. However, the above studies have some common results that are of relevance to the accurate prediction of reactivity. First, the prediction of ozone concentrations will have less of a mechanism-related bias than other species such as HCHO, and organic peroxides, which are more heavily dependent on the details of the chemical mechanism. This does not imply that the mechanisms are correct, but does imply that the ozone predictions resulting from the use of different mechanisms may be similar. Second, some of the observed differences may be due to differences in the HO 2 and RO 2 reactions used. The use of pressure and H 2 O dependant rates for the HO 2 self-reaction has a significant effect on the HO x budget in the simulations. Third, variations in the treatment of photolysis for clear sky conditions may affect O 3 mixing ratios, but the effects of the different treatments have a similar magnitude to the measurement errors in the input data used in the models. Increased precision in these measurements is required to improve the estimates of photodissociation rates.finally, mechanisms with increased speciation or production of oxygenated hydrocarbons such as organic peroxides and aldehydes tended to have higher ozone production than other mechanisms (Kuhn et al., 1998). However, this does not necessarily imply that these species will necessarily be in abundance in the real atmosphere. Deposition and particle formation may result in lower mixing ratios of these species than would be predicted by gas-phase chemistry alone box models as employed in the above studies. In addition, the kinetics and mechanisms of many oxygenated species have yet to be determined in the laboratory; current detailed mechanisms extrapolate from known chemistry. The kinetics and atmospheric fate of the oxygenated compounds are poorly known relative to the unoxygenated species and are worthy of further study. Accurate predictions of these species are a much more stringent test of the accuracy of a model s VOC oxidation pathways than the model s predictions of ozone mixing ratios. CURRENT STATUS FOR REACTIVITY MODELING Given below is a brief summary of the status and updates to the latest version of the SAPRC mechanism for the various major classes of compounds, and the results of the evaluation 19

35 of those mechanisms, where applicable. Although this discussion is strictly speaking applicable only to that mechanism, it is probably reasonably representative of the current state of the science of reactivity assessment for various classes of compounds. Alkanes. Atkinson (unpublished results) has obtained new product yields for alkyl nitrates from C 5 C 10 n-alkanes indicating that the previously published yields in these systems may be high by about 30%. When the nitrate yields for the higher alkanes are reduced accordingly, it is now possible to fit the chamber data for the C 8 + n-alkanes without making the unreasonable assumption that nitrate formation does not occur from the peroxy radicals formed after 1,4-H shift isomerizations. The estimated mechanism gave generally satisfactory fits to reactivity data for most alkanes except for iso-octane (2,2,4-trimethylpentane), where some adjustments were necessary. There may be a tendency for the mechanism to overpredict the inhibition by the higher alkanes in the mini-surrogate runs, but it is unclear whether this is a consistent bias. Mineral Spirits. The individual branched and cyclic alkane isomers which were used to represent the various classes of the alkane mineral spirits which were studied (Carter et al., 1997b) were modified somewhat based on the analyses supplied by Safety-Kleen corporation. Their analysis indicated that the mixtures are characterized by somewhat less highly branched isomers than we had been assuming previously. With this change, and the change in estimated nitrate yields in the general alkane mechanism as discussed above, the model could successfully simulate the results of the mineral spirits reactivity experiments without adjustment. Additional data from our ongoing programs will be needed to confirm this. However, the uncertainty classifications for the higher branched and cyclic alkane classes have been reduced based on these current results. Alkenes. The automated mechanism generation procedure used with the SAPRC-98 mechanism allows for more realistic and complex mechanisms to be generated for the higher alkanes, though it is still assumed that all the reaction with OH radicals is by addition to the double bond. However, the evaluations of the mechanisms for the simpler alkenes (whose mechanisms are not significantly affected by the use of this automated procedure) indicate problems and inconsistencies that have not been satisfactorily been resolved. In particular, in order to fit chamber data for 1-butene and 1-hexene, it is necessary to assume lower OH radical yields in the reactions of O 3 with these compounds than is consistent with recommendations of Atkinson (1997) based on results of various other laboratory studies. In fact, the previous version of the mechanism also performed poorly in simulating experiments with these compounds, though this had not been recognized until this re-evaluation. It is also necessary to assume that essentially no radicals are formed in the reactions of O( 3 P) with C 3 + monoalkenes, contrary to the assumptions of previous models. On the other hand, the isoprene data are still best fit if the relative high radical yields in the O 3 and O( 3 P) reactions of this compound are assumed, and the terpene data are also reasonably well fit using the recommended (generally relatively high) OH yields in their O 3 reactions. Although the mechanisms for the various alkenes were adjusted if needed to fit the available chamber data, their mechanisms must be considered to be somewhat uncertain until these inconsistencies are resolved. 20

36 Aromatics. Despite considerable research in recent years and some progress, the details of the aromatic ring opening process is remains sufficiently poorly understood that use of parameterized and adjusted mechanisms is still necessary. Some changes were made to the details of the parameterization to permit use in the model of the actual observed dicarbonyl products, but the general parameterization approach was the same. The parameters were optimized to fit the chamber data for the various compounds for which data are available, and the fits to the chamber data were comparable (though usually slightly better) to those for the SAPRC-97 mechanism (Carter et al., 1997). The approach for representing the higher aromatics in the model was also modified somewhat. Ethylbenzene, which was found to have a lower mechanistic reactivity than toluene, was used rather than toluene to represent the higher monoalkylbenzenes. The generic di- and tri- or polyalkylbenzenes were represented by mixtures of xylene or trialkylbenzene isomers, rather than just m-xylene or 1,3,5-trimethylbenzene, as was the case previously. This was done to eliminate a potential source of bias in the mechanism by representing each of these classes by what is essentially the most reactive member of the class Ketones. The previous mechanism used methylethyl ketone (MEK) to represent essentially all ketones other than acetone, and chamber data with methyl isobutyl ketone (MIBK) has shown this to be unsatisfactory. The current mechanism represents individual ketones based on their estimated individual reactions, generated as discussed above for alkanes and other oxygenates. This has resulted in significant changes in predicted reactivities for the higher ketones. Other Oxygenated Species. The estimated mechanisms of alcohols, ethers, glycols, esters, etc. were generated using the automated procedure discussed above, with nitrate yields and other uncertain branching ratios in the mechanism being adjusted to fit chamber data if necessary. Footnotes in the reactivity data tabulations indicate the types of adjustments that were made, if necessary. All the chamber data obtained in the just-completed CARB consumer products reactivity program have been utilized and incorporated in developing and adjusting these mechanisms (Carter et al., unpublished results). The chamber data for most of these compounds could be fit with the model simulations after adjustments that were considered to be within the uncertainty range of the estimation method. A very preliminary analysis of the differences in reactivities between the initially estimated mechanisms and those adjusted to fit chamber data suggests that the MIR differences are usually less than a factor of 1.5, though a more complete analysis is needed. Based on this preliminary analysis, we conclude that at least for C 8 - oxygenated compounds, the generated using the estimated mechanism are probably not off by more than a factor of two. Halogenated Compounds. With the exception of chloropicrin, which appears to have relatively simple and unique chemistry (Carter et al., 1997c), the few halogenated compounds we have studied (trichloroethylene and alkyl bromides) indicate that current mechanisms cannot reliably predict MIR s for these compounds (Carter et al., 1996, 1997d). The current version of the mechanism does not yet include halogen chemistry, though chlorine and bromine reactions may be added before the mechanism is finalized. Nitrogen Containing Compounds. The only nitrogen-containing compound with an evaluated mechanism in the current list is n-methyl pyrrolidinone (NMP). For the simple amines for which OH radical rate constants are available, highly simplified placeholder mechanisms 21

37 with the appropriate OH rate constant, are used to provide best estimate reactivities. These must be considered to be highly uncertain. Other Organic Classes. Although data area available for toluene diisocyanate and some volatile silicone compounds, these compounds, which appear to inhibit ozone under most if not all conditions, are not yet represented in the current mechanism. Other classes of compounds, which were represented by highly approximate placeholder mechanisms, are also omitted from this version of the mechanism. CONCLUSIONS There has been significant progress in recent years in improving our understanding and ability to model the gas-phase reactions of pollutants in the troposphere. Chemical mechanisms have been or are being developed which are able to predict ozone impacts in simulated atmospheric conditions for many of the types of organic compounds that are emitted into the atmosphere, particularly those emitted in large amounts. Nevertheless, there remain major gaps in our understanding in the details of these gas-phase reactions, and our understanding of potentially significant heterogeneous processes is even more incomplete. Many of the mechanisms used to predict ozone are highly parameterized and simplified, with empirical adjustments to fit chamber data, and with no reliable ability to predict impacts other than on ozone and perhaps overall OH radical levels. Although methods exist for estimating mechanisms for VOCs for which there are no data, at best these estimates have large uncertainties, and for many classes of compounds no reliable estimation methods exist Ongoing improvements in airshed model hardware and software is permitting use of ever more detailed atmospheric mechanisms, which have the potential to give more accurate and comprehensive atmospheric impact predictions for any VOC of interest. However, without the knowledge of the mechanistic details, the predictions using these detailed mechanisms may be no more reliable than those of the simplified and parameterized mechanisms currently in use. Therefore, the main factor limiting the chemical accuracy of current state-of-the-art and future airshed models is limitations in our knowledge of atmospheric chemistry, and limitations in the environmental chamber data base needed to verify the accuracy of the chemical model predictions. Specific areas where research is needed to improve the chemical mechanisms needed for VOC reactivity assessment are as follows: Basic kinetic and mechanistic studies are needed to improve our understanding and to reduce uncertainties concerning the atmospheric reactions of many types of VOCs, in particular with regard to the reactions of the intermediate radicals and reactive products they form. Atkinson (1999) has suggested a number of areas where work is needed, and additional areas are discussed in this document. These include, but are not limited to, areas which have traditionally been recognized as uncertainties in atmospheric chemistry, such as mechanism for reactions of aromatics, the ozone olefin reactions, organic nitrate yields from peroxy + NO reactions, relative rates of the many competing alkoxy radical reactions, photolysis of oxygenated product 22

38 species (particularly from aromatics), reducing the uncertainty in the cross-sections and quantum yields of other species, including inorganics, etc. Our understanding and ability to predictively model the heterogeneous reactions in the atmosphere and in environmental chambers needs to be improved. This includes, but is not limited to, reactions involving HONO formation, N 2 O 5 hydrolysis, and reactions involved in secondary formation, including secondary organic aerosols. Laboratory and ambient air measurements are required to determine the atmospheric fate of potentially reactive oxygenated species, particularly for those with a high OH reactivity (or expected high OH reactivity) and a known or expected low volatility. Better understanding of reactions in smog chambers is needed to reduce uncertainties and sources of errors when using environmental chamber data to evaluate and develop chemical mechanisms. Issues brought up in the NARSTO review of heterogeneous processes in aqueous media by Daniel Jacob (1999) need to be considered. There needs to be improvement in the environmental chamber methodologies used to develop and evaluate mechanisms for atmospheric models. Improved facilities are needed to evaluate mechanisms under lower pollutant conditions than is currently possible, and improved instrumentation is needed to monitor trace species and intermediates and VOC reaction products. This is needed for more comprehensive and reliable mechanism evaluation, particularly for impacts other than ozone, and improved confidence that the mechanism will give correct predictions under ambient conditions. On the other hand, methodologies need to be developed to screen or assess reactivity more readily and at lower cost than is currently is possible. In addition to the obvious practical benefits in aiding implementation of reactivity-based control strategies, such methodologies will provide valuable data needed to improve and evaluate estimation methods. Of critical importance to developing improved chemically detailed mechanisms is developing improved and more reliable estimation techniques. This provides the only practical means for developing and implementing fully detailed mechanisms for the full variety of VOCs of interest in the foreseeable future. Structure-reactivity methods such as those developed by Atkinson and others have proven to be powerful tools, but the theoretical and experimental data base limits their utility to restricted classes, and many estimates are uncertain. Theoretical calculations of the most uncertain reactions and targeted experimental studies to provide needed data to establish or evaluate relationships are needed. A serious sensitivity analysis should be applied to the mechanisms to help decide which processes will be most important to study in the future. This requires quantifying the uncertainties involved, not only in the elementary rate constants, but also in parameterization methods in mechanisms adjusted to fit chamber or other data. Progress is being made in this area by Milford and others, but the results are still of limited utility. Although computer hardware and software is improving the level of chemical detail that can be represented in models, fully explicit and complete mechanisms are not now and probably will never be practical. Initial numerical intercomparisons between chemical mechanisms of very different levels of complexity have shown relatively small variations in ozone predictions compared to other species Work is needed to assess the optimum level of detail for atmospheric chemical mechanisms, given the modeling application and the level of knowledge of the 23

39 processes being represented. The creation of a minimum list of inorganic reactions required for photochemical reactivity calculations should be formed, which includes the pressure and water vapor dependent pathways of HO 2. A model comparison of highly speciated versus lumped versus temporarily compressed mechanisms should be performed for a realistic atmospheric conditions, to determine the relative merits of model speciation in reactivity estimates. The implementation of the Morphecule approach needs to be completed, and its advantages over alternative methods for representing chemical detail in models need to be assessed. REFERENCES Andersson-Skold, Y., Grenfelt, P., and Pleije, K. (1992): J. Air Waste Mgmt. Assoc., 42: Ammann, M. and R. Lorenzen (Editors) Proceedings of the 1 st workshop of the EUROTRAC-2 Subproject Chemical Mechanism Development, PSI-Proceedings 97-02, Paul Scherrer Institut, 5232 Villigen PSI, Switzerland, Ammann, M., M. Kalberer, D. T. Jost, L. Tobler, E. Rossler, D. Piguet, H. W. Gaggeler, and U. Baltensperger (1998): "Heterogeneous production of nitrous acid on soot in polluted air masses," Nature, 395, Andres-Hernandez, M.D., J. Notholt, J. Hjorth and O. Schrems, A DOAS study on the origin of nitrous acid at urban and non-urban sites, Atm. Env., 30, , Atkinson, R. (1999): Atmospheric Chemistry of VOCs and NOx NARSTO Assessment Critical Review, Atmos. Environ., in press. (This can be downloaded from Atkinson, R. and W. P. L. Carter (1984): Kinetics and Mechanisms of the Gas-Phase Reactions of Ozone with Organic Compounds under Atmospheric Conditions, Chem. Rev. 1984, Atkinson, R. (1988): Gas-phase atmospheric chemistry of organic compounds, Final Report to the California Air Resources Board, Contract No. A , Sacremento, Ca. Atkinson, R. (1989): Kinetics and Mechanisms of the Gas-Phase Reactions of the Hydroxyl Radical with Organic Compounds, J. Phys. Chem. Ref. Data, Monograph no 1. Atkinson, R. (1990): Gas-Phase Tropospheric Chemistry of Organic Compounds: A Review, Atmos. Environ., 24A, Atkinson, R. (1991): Kinetics and Mechanisms of the Gas-Phase Reactions of the NO 3 Radical with Organic Compounds, J. Phys. Chem. Ref. Data, 20, Atkinson, R. (1994): Gas-Phase Tropospheric Chemistry of Organic Compounds, J. Phys. Chem. Ref. Data, Monograph No

40 Atkinson, R. (1997): Gas Phase Tropospheric Chemistry of Volatile Organic Compounds: 1. Alkanes and Alkenes, J. Phys. Chem. Ref. Data, 26, Atkinson, R., D. L. Baulch, R. A. Cox, R. F. Hampson, Jr., J. A. Kerr, M. J. Rossi, and J. Troe (1997): Evaluated Kinetic, Photochemical and Heterogeneous Data for Atmospheric Chemistry: Supplement V., IUPAC Subcommittee on Gas Kinetic Data Evaluation for Atmospheric Chemistry, J. Phys. Chem. Ref. Data, 26, Bambauer, A., B. Brantner, M. Paige, and T. Novakov (1994): Atmos. Environ., 28, Becker, K.H., The atmospheric oxidation of aromatic hydrocarbons and its impact on photooxidant chemistry, in Proceedings of Eurotrac Symposium 94, Editors P.M. Borrel, P. Borrel, T. Cvitas, and W. Seilev, SPB Academic Publishing bv, The Hague, Netherlands, 66-74, Benson, S. W. (1976): Thermochemical Kinetics, 2nd Ed., John Wiley and Sons, New York. Bergin, M. S., A. G. Russell, W.P.L. Carter, B. E. Croes and J. H. Seinfeld (1997): Ozone Control and VOC Reactivity, in Encyclopedia of Environmental Analysis and Remediation, Wiley. Brown et al.1998, S. S., Talukdar, R.K. and Ravishankara, A. R. (1998): Chem. Phys. Lett. (submitted) CARB (1993): Proposed Regulations for Low-Emission Vehicles and Clean Fuels Staff Report and Technical Support Document, California Air Resources Board, Sacramento, CA, August 13, See also Appendix VIII of California Exhaust Emission Standards and Test Procedures for 1988 and Subsequent Model Passenger Cars, Light Duty Trucks and Medium Duty Vehicles, as last amended September 22, Incorporated by reference in Section (k) of Title 13, California Code of Regulations. Carter, W.P.L. (1988): Documentation of a gas phase photochemical mechanism for use in airshed modeling, Final Report to the California Air Resources Board, Contract No. A , California Air Resources Board, Sacremento, Ca. Carter, W. P. L. (1990): A Detailed Mechanism for the Gas-Phase Atmospheric Reactions of Organic Compounds, Atmos. Environ., 24A, Carter, W. P. L. (1994): Development of Ozone Reactivity Scales for Volatile Organic Compounds, J. Air & Waste Manage. Assoc., 44, Carter, W. P. L. (1995): Computer Modeling of Environmental Chamber Measurements of Maximum Incremental Reactivities of Volatile Organic Compounds, Atmos. Environ., 29,

41 Carter, W.P.L., F.W. Lurmann, R. Atkinson and A.C. Lloyd (1986): Development and testing of a surrogate species chemical reaction mechanism, Volume 1, PB /AS, National Technical Information Service, Springfield, Va. Carter, W. P. L. and R. Atkinson (1989): A Computer Modeling Study of Incremental Hydrocarbon Reactivity, Environ. Sci. Technol., 23, 864. Carter, W. P. L., and F. W. Lurmann (1990): Evaluation of the RADM Gas-Phase Chemical Mechanism, Final Report, EPA-600/ Carter, W. P. L. and F. W. Lurmann (1991): Evaluation of a Detailed Gas-Phase Atmospheric Reaction Mechanism using Environmental Chamber Data, Atm. Environ. 25A, Carter, W. P. L, D. Luo, I. L. Malkina, and J. A. Pierce (1993): An Experimental and Modeling Study of the Photochemical Ozone Reactivity of Acetone, Final Report to Chemical Manufacturers Association Contract No. KET-ACE-CRC-2.0. December 10. (This can be downloaded from bycarter.htm) Carter, W. P. L., D. Luo, I. L. Malkina, and D. Fitz (1995b): The University of California, Riverside Environmental Chamber Data Base for Evaluating Oxidant Mechanism. Indoor Chamber Experiments through 1993, Report submitted to the U. S. Environmental Protection Agency, EPA/AREAL, Research Triangle Park, NC., March 20. (This can be downloaded from bycarter.htm) Carter, W. P. L., D. Luo, and I. L. Malkina (1996): Investigation of the Atmospheric Ozone Formation Potential of Trichloroethylene, Report to the Halogenated Solvents Industry Alliance, August. (This can be downloaded from bycarter.htm.) Carter, W. P. L., D. Luo, and I. L. Malkina (1997): Environmental Chamber Studies for Development of an Updated Photochemical Mechanism for VOC Reactivity Assessment, final report to California Air Resources Board Contract , Coordinating Research Council Project M-9, and National Renewable Energy Laboratory Contract ZF November 26. (This report can be downloaded from Carter, W. P. L., D. Luo, and I. L. Malkina (1997b): Investigation of the Atmospheric Ozone Formation Potentials of Selected Mineral Spirits Mixtures, Report to Safety-Kleen Corporation, July 25.. (This report can be downloaded from ~carter/bycarter.htm.) Carter, W. P. L., D. Luo and I. L. Malkina (1997c): Investigation of that Atmospheric Reactions of Chloropicrin, Atmos. Environ. 31,

42 Carter, W. P. L., D. Luo, and I. L. Malkina (1997c): Investigation of the Atmospheric Ozone Formation Potential of Selected Alkyl Bromides, Report to Albemarle Corporation, Novemver 10.. (This can be downloaded from Derwent, R.G., M.E. Jenkin and S.M. Saunders (1996): Photochemical ozone creation potentials for a large number of reactive hydrocarbons under European conditions, Atmos. Environ. 30,189. Derwent, R. G. and M.E. Jenkin (1991): Hydrocarbons and the long range transport of ozone and PAN across Europe, Atmos. Environ. 25A Dodge M.C., (1999): Chemical oxidant mechanisms for air quality modeling: critical review paper for 1998 ozone assessment (to be submitted to Atmospheric Environment Available at Dransfield et al. 1998, T. J., Sprengnether, M. M., Perkins, K. K., Donahue, N. M. and Anderson, J. G. (1998) Geophys. Res. Lett. (submitted) Gerecke, A., A. Thielmann, L. Gutzwiller and M.J. Rossi, The chemical kinetics of HONO formation resulting from heterogeneous interaction of NO 2 with flame soot, Proceedings of the 1 st workshop of the EUROTRAC-2 Subproject Chemical Mechanism Development, Editors M. Amman and R. Lorenzen, 23-26, PSI-Proceedings 97-02, Paul Scherrer Institut, 5232 Villigen PSI, Switzerland, Gery, M. W., G. Z. Whitten, and J. P. Killus (1988): Development and Testing of the CBM-IV For Urban and Regional Modeling,, EPA-600/ , January. Gery M.W., G. Z. Whitten, J.P. Killus, and M.C. Dodge, A photochemical kinetics mechanism for urban and regional scale computer modeling, J. Geophys. Res, 94, , Gierczak, T., J.B. Burkholder, R.K. Talukdar, A. Mellouki, S.B. Barone, A.R. Ravishankara, "Atmospheric fate of methyl vinyl ketone and methacrolein," J. Photochem. and Photobio. A: Chem., 1-10, Hobson, J. F., R. Atkinson and W. P. L. Carter (1997): Volatile Methylsiloxanes, in Handbook of Environmental Chemistry, Vol. 3, Part H: Organosilicon Materials, Springer-Verlag. Horie, O., P. Neeb, S. Limbach and G.K. Moortgat, Formation of formic acid and organic peroxides in the ozonolysis of ethene with added water vapour, Geophys. Res. Lett., 21, , 1994 Iraci, L. T., and M. A. Tolbert (1997): "Heterogeneous interaction of formaldehyde with cold sulfuric acid: Implications for the upper troposphere and lower stratosphere," J. Geophys. Res., 102, 16,099-16,

43 Irikura, K. K., and D. J. Frurip, Editors (1998): Computational Thermochemistry; ACS Symposium Series #677 Jacob, D (1999): Heterogeneous Chemistry and Tropospheric Ozone, Critical review paper for NARSTO, Atmospheric Environment, in press. Available at Narsto/assess_activities.html. Jacobson, M.Z., and R.P. Turco, SMVGEAR: A sparse-matrix, vectorized Gear Code for Atmospheric Models, Atm. Env., 28, , 1994 Jeffries, H. E., M. W. Gery and W. P. L. Carter (1992): Protocol for Evaluating Oxidant Mechanisms for Urban and Regional Models, Report for U. S. Environmental Protection Agency Cooperative Agreement No , Atmospheric Research and Exposure Assessment Laboratory, Research Triangle Park, NC. Jeffries, H. E., R. M. Kamens, K. G. Sexton, and A. A. Gerhardt (1982): Outdoor Smog Chamber Experiments to Test Photochemical Models, EPA-600/ a, April. Jenkin, M.E., S.M. Saunders and M.J. Pilling (1997): The tropospheric degradation of volatile organic compounds: a protocol for mechanism development,. Atmos. Environ. 31, 31. Kamm, S., O. Mohler, K.-H. Naumann, H. Saathoff, U. Schurath, Temperature dependence of slow heterogeneous reactions on soot aerosol, Proceedings of the 1 st workshop of the EUROTRAC-2 Subproject Chemical Mechanism Development, Editors M. Amman and R. Lorenzen, 29-31, PSI-Proceedings 97-02, Paul Scherrer Institut, 5232 Villigen PSI, Switzerland, Kelly, N.A.; Wang, P.; Japar, S.M.; Hurley, M.D.; and Wallington, T.J. (1994). Measurement of the Atmosphere Reactivity of Emissions from Gasoline and Alternative-Fueled Vehicles: Assessment of Available Methodologies, First-Year Final Report, CRC Contract No. AQ , Environmental Research Consortium. Kelly, N.A. and Wang, P. (1996) Part I: Indoor Smog Chamber Study of Reactivity in Kelly, N.A.; Wang, P.; Japar, S.M.; Hurley, M.D.; and Wallington, T.J. (1996). Measurement of the Atmosphere Reactivity of Emissions from Gasoline and Alternative-Fueled Vehicles: Assessment of Available Methodologies, Second-Year Final Report, CRC Contract No. AQ and NREL Contract No. AF Environmental Research Consortium, - (September). Kleffmann, J., K.H. Becker and P. Wiesen, Mechanisms of heterogeneous HONO formation in urban areas, Proceedings of the 1 st workshop of the EUROTRAC-2 Subproject Chemical Mechanism Development, Editors M. Amman and R. Lorenzen, 32-34, PSI-Proceedings 97-02, Paul Scherrer Institut, 5232 Villigen PSI, Switzerland,

44 Kuhn, M., P.J.H. Builtjes, D. Poppe, D. Simpson, W.R. Stockwell, Y. Andersson-Skold, A. Baart, M. Das, F. Fiedler, O. Hov, F. Kirchner, P.A. Makar, J.B. Milford, M.G.M. Roemer, R. Ruhnke, A. Strand, B. Vogel, and H. Vogel, Intercomparison of the gasphase chemistry in several chemistry and transport models, Atm. Env., 32, , Kwok, E. S. C., and R. Atkinson (1995): Estimation of Hydroxyl Radical Reaction Rate Constants for Gas-Phase Organic Compounds Using a Structure-Reactivity Relationship: An Update, Atmos. Environ 29, Lane, D.A., S.S. Fielder, S.J. Townsend, N.J. Bunce, J. Zhu, L. Liu, B. Wiens, and P.Pond, Atmospheric photochemistry of napthalene: A practical and theoretical approach, Polycyclic Aromatic Compounds, 9, 53-59, Lurmann, F.W., A.C. Lloyd, R. Atkinson (1986): A chemical mechanism for use in long-range transport/acid deposition computer modeling, J. Geophys. Res., 91, Lurmann, F.W., W.P.L. Carter and L.A. Coyner (1987): A surrogate species chemical reaction mechanism for urban-scale air quality simulation models: Volume 1. Adaptation of the mechanism, PB /AS, National Technical Information Service, Springfield, Virginia. Lurmann, F. W., M. Gery, and W. P. L. Carter (1991): Implementation of the 1990 SAPRC Chemical Mechanism in the Urban Airshed Model, Final Report to the California South Coast Air Quality Management District, Sonoma Technology, Inc. Report STI FR, Santa Rosa, CA. Makar, P.A., S.-M. Li, P.B. Shepson, and J. Bottenheim (1998): The AES gas-phase mechanism for tropospheric chemistry: theoretical formulation, AES Internal Report, Atmospheric Environment Service, 4905 Dufferin Street, Downsview, Ontario, Canada. Makar, P.A., J.D. Fuentes, D. Wang, R.M. Staebler, H.A. Wiebe (1998b): Chemical processing of biogenic hydrocarbons within and above a temperate deciduous forest, J. Geophys. Res. (accepted; in press). Makar, P.A., H.A. Wiebe, R.M. Staebler, S.M. Li and K. Anlauf (1998c): Measurement and modeling of particle nitrate formation, J. Geophys. Res., 103, Makar, P.A. (1998): The use of transient mechanism compression to reduce chemical integration processing time, Proceedings, International Conference on Air Pollution Modelling and Simulation (APMS 98), (also currently under review for journal special issue on conference). Makar, P.A., W.R. Stockwell, and S.M. Li (1996): Gas-phase chemical mechanism compression strategies: treatment of reactants, Atm. Env., 30,

45 Makar, P.A. and S.M. Polavarapu (1997): Analytic solutions for gas-phase chemical mechanism compression, Atm. Env., 31, Middleton, P, W.R. Stockwell, and W.P.L. Carter, Aggregation and analysis of volatile organic compound emissions for regional modelling, Atm. Env, 24, , Moran, M.D., A. Dastoor, S.-L. Gong, W. Gong, P.A. Makar, Conceptual design for the AES regional particulate-matter model/unified air quality model, AES Internal Report, Air Quality Modelling and Integration Division, 4905 Dufferin St., Downsview, Ontario, Canada, 100 pp., NASA (1997): Chemical Kinetics and Photochemical Data for Use in Stratospheric Modeling, Evaluation Number 12, JPL Publication 97-4, Jet Propulsion Laboratory, Pasadena, California, January. Odum, J.R.; Hoffmann, T.; Bowman, F.; Collins, D.; Flagan, R.C.; and Seinfeld, J.H. (1996). Environ. Sci. Technol., 30: Olson, J., M. Prather, T. Berntsen, G. Carmichael, R. Chatfield, P. Connell, R. Derwent, L. Horowitz, S. Jin, M. Kanakidou, P. Kasibhatla, R. Kotamarthi, M. Kuhn, K. Law, J. Penner, L. Perliski, S. Sillman, F. Stordal, A. Thompson, O. Wild, Results from the Intergovernmental Panel on Climate Change photochemical model intercomparison (PhotoComp), J. Geophys. Res., 102, , Stockwell, W.R., F. Kirchner, M. Kuhn, and S. Seefeld, A new mechanism for regional atmospheric chemistry modeling. J. Geophys. Res., 102, , Stockwell, W. R., P. Middleton, J. S. Chang, and X. Tang (1990): The Second Generation Regional Acid Deposition Model Chemical Mechanism for Regional Air Quality Modeling, J. Geophys. Res. 95, Stockwell, W.R. and F.W. Lurmann, Intercomparison of the ADOM and RADM gas-phase chemical mechanisms, Electric Power Research Institute Topical Report, EPRI, Palo Alto, Ca, Tuazon, E. C., S. M. Aschmann, R. Atkinson and W. P. L. Carter (1998): The Reactions of Selected Acetates with the OH Radical in the Presence of NO: Novel Rearrangement of Alkoxy Radicals of Structure RC(O)OCH(O)R, J. Phys. Chem. 102, Wang, S.C.; Flagan, R.C.; and Seinfeld, J.H. (1992). Atmos. Environ., 26A, Williams, L. R. and Golden, D. M. (1998) Evaluation prepared for the next JPL Panel Report. 30

46 EMISSIONS DATA 5/5/99 P.A. Makar 1 (lead), M.D. Moran 1, A. Russell 2, G. Sistla 3 1 Atmospheric Environment Service,, Ontario, Canada 2 Georgia Institute of Technology, Atlanta 3 New York State Department of Environmental Conservation CONTENTS OVERVIEW AND SUMMARY OF RECOMMENDATIONS...2 References...5 EMISSIONS UNCERTAINTIES AND IMPACT ON REACTIVITY ASSESSMENTS...6 References...7 STATUS OF USE OF ANTHROPOGENIC EMISSIONS IN AIR QUALITY ASSESSMENTS...8 Stationary Sources...8 Point and Area sources...9 Mobile sources...10 Speciation...10 Future work...11 References...11 BIOGENIC EMISSIONS AND CHEMICAL INTERACTIONS...13 References

47 OVERVIEW AND SUMMARY OF RECOMMENDATIONS P.A. Makar and M.D. Moran The emissions from the multitude of sources in the environment are critically important inputs in the airshed models that must be used to assess impacts or reactivities of VOCs. The amount of emissions determines the nature of the chemical environment in which the VOC reacts, and thus affects not only the absolute amount of ozone and other secondary pollutants formed, but also the incremental effects of any additional VOC on the formation of those pollutants (i.e., the VOC reactivities). Emissions data also impact directly on government policy. Changes in emissions levels over time reflect the effectiveness of government regulations, as well as changes in activity level. The accuracy and completeness of emissions databases are therefore of great importance. The creation of an emissions database takes place in several stages, with each stage having associated errors and uncertainties. From sources such as Moody et al (1995), the SPECIATE website (Ryan, 1998), Moran et al (1997), and the AP-42 database (EPA, 1995), typical stages for determining anthropogenic emissions are as follows: Source profiles (which provide a breakdown of the total VOC emitted from a given source type into a detailed VOC fractionation) are constructed from measurements. A limited number of profiles are currently available (e.g. about 300 used in SPECIATE). Source types are categorized using Source Classification Codes (SCC), Area Mobile Source codes (AMS) and Standard Industrial Classification codes (SIC). Typical emissions processing systems have thousands of source classifications (Ryan, 1998). Each classified source type is assigned a source profile from the set of available profiles. The total annual VOC emission for each source type is determined from the product of the appropriate emission factor, activity rate, and control factors (AP-42; EPA, 1995). Individual sources from the database are assigned total annual VOC emissions, source classification codes, and source profiles. The total annual VOC emission rates may be multiplied by meteorological factors (e.g. to include dependence of emissions on ambient temperatures). The source profiles, linked to the individual sources in (5), are used to convert the total annual VOC emission into a speciated annual VOC emission. The detailed speciation is lumped into a smaller number of species for modelling purposes (e.g. Middleton et al., 1990). Temporal allocation occurs; temporal factors are applied to the model species annual emissions to create seasonal, daily, then hourly emissions estimates. 2

48 Spatial allocation factors are used to create the spatial distribution of area sources. Smaller point sources are sometimes included with the area sources. For biogenic emissions, the processing differs somewhat: Spatially varying biomass factors (mass of emitting foliage / unit horizontal area) are linked with monthly frost/seasonal variation factors and a database of vegetation categories. Net vegetation standard emission rates (30 o C, 1000 µmol/m 2 /sec photo-synthetically active radiation (PAR)) are calculated for each part of the vegetation grid. Temperature and / or PAR corrections are applied for the meteorology at the site. For both forms of emissions, regridding from the database spatial grid to the model grid is often required. Each of the stages listed above has uncertainties, many of which have not been characterized in a quantitative way. The limited number of source profiles compared to SCC, SIC, or AMS codes is an example. The extent to which the available source profiles adequately represent the much larger set of source classifications is unknown. The source profiles themselves are the result of a relatively small number of measurement studies; in some cases a single study is used to represent the emissions from an entire industry. Errors in the total annual VOC emission rates are usually not estimated. Similarly, errors in temporal emission factor estimates are usually not estimated. For the biogenic emissions, the existence of detailed vegetation databases is required; these are not always available with sufficiently fine spatial detail for adequate resolution of emitting regions. Some of the sources of uncertainty have been characterized. The effect of lumping detailed speciation emissions into model speciation has been examined by Middleton et al (1990) - comparisons there and in later work (Kuhn et al., 1998; see also Lumping in the Atmospheric Chemistry chapter of this assessment) indicate that the effects of speciation in model ozone production may be a smaller source of error than other model approximations. Similarly, the procedures for determining standard emission rates for specific vegetation types are well established and the uncertainties associated with these measurements are known (c.f. Guenther et al., 1996). From the standpoint of VOC reactivity and ozone formation, it is essential to determine which emissions-related uncertainties have the biggest impact on the predicted reactivity of the atmosphere. This in turn could be used to design measurement initiatives directed at reducing the most significant uncertainties. Sensitivity studies of model results towards errors in the temporal factors, source profile assignment and total VOC emission rates should be directed towards determining which aspects of the emissions system have the biggest impact on reactivity. A few preliminary studies of this nature are discussed in the Emissions Uncertainty section, below. Emissions uncertainties can contribute 30 to 50% of the overall uncertainty in O3 predictions, while the contribution to uncertainty in VOC reactivity may be smaller (8%). 3

49 Certain aspects of anthropogenic emission factor calculation are described in the Status of Anthropogenic Emissions section, below. There it is noted that all available data for point sources should be taken into account, for both minor and major emitters. In addition, time factors and other parameters have typically been estimated based on episode days; non-episode, nonattainment days may therefore be incorrectly represented in emissions databases. Biogenic hydrocarbon emissions are described in a separate section, below, with the need for improved vegetation databases noted as the most significant need for further research. Emission reduction effects such as volatility suggest that some AP-42 emissions estimates may not adequately describe the amounts of VOCs that ultimately enter the atmosphere. These arguments are presented separately in the Emissions and Volatility chapter of this assessment, since they are relevant to assessing reactivities of individual VOCs as well as the quality of the overall emissions inventory. Overall, the following recommendations can be made based on the work outlined above and in the subsequent sections: There is a clear need to determine the impact of the approximations used in the emissions assignment process on the results of models used for reactivity calculations. Sensitivity analyses should be conducted to identify the effect of approximations in temporal factors, source profile assignment and total annual VOC emission rates on model reactivity estimates. Very limited research to date suggests that, at least in the Southern California area, emissions uncertainties have a more limited impact on our ability to quantify VOC relative reactivities. In part, this makes sense since the addition of more VOC tends to reduce the reactivities of all VOCs, such that the relative reactivity of one species to another is not so impacted. However, the robustness of this conclusion is not yet demonstrated. For example, the uncertainties in the biogenic emissions are large, and may have a more marked effect in domains like the eastern United States. Given the continued, and seeming perpetual existence, of large uncertainties in emissions, particularly biogenics, this issue should be explored further. The relative impact of emissions uncertainties on both absolute O 3 mixing ratios and on the estimated VOC reactivity should be examined in sensitivity analyses. The impact on O 3 is expected to be larger than on relative reactivities, but this should be confirmed in simulations for different airshed conditions. There is a need for the improvement of the vegetation land-use database and area average factors used in biogenic emissions estimates to reduce their uncertainty. Other significant data gaps include lack of factors genetic variability within species and for wounding of vegetation, inadequate characterization of emission underestimates due to chemical losses, and the need for refinement of emission activity and escape efficiency parameters. EIIP procedures, operational information, and stack sampling should be used to ensure adequate speciation of anthropogenic VOCs, with the encouragement of industry participation. 4

50 Time factors for apportionment of VOCs on non-episode days need to be designed, as well as appropriate changes to engineering and VOC composition parameters to reflect non-episode emissions days. Updates to the limited set of speciation profiles are essential for improvement to the emissions databases. Volatility factors need to be designed to account for the effects of volatility on net emissions. This is discussed further in the Volatility and Fate chapter of this assessment. References AP-42 database on Air Chief 4.0 CD-ROM, EFIG/EMAD/OAQPS/EPA, USEPA, 1995 Guenther, A., W. Baugh, K. Davis, G. Hampton, P. Harley, L. Klinger, L. Vierling, P. Zimmerman, E. Allwine, S. Dilts, B. Lamb, H. Westburg, D. Baldocchi, C. Geron and T. Pierce, Isoprene fluxes measured by enclosure, relaxed eddy accumulation, surface layer gradient, mixed layer gradient, and mixed layer mass balance techniques, J. Geophys. Res., 101, , Kuhn, M., P.J.H. Builtjes, D. Poppe, D. Simpson, W.R. Stockwell, Y. Andersson-Skold, A. Baart, M. Das, F. Fiedler, O. Hov, F. Kirchner, P.A. Makar, J.B. Milford, M.G.M. Roemer, R. Ruhnke, A. Strand, B. Vogel, and H. Vogel, Intercomparison of the gasphase chemistry in several chemistry and transport models, Atm. Env., 32, , Middleton, P., W.R. Stockwell, and W.P.L. Carter, Aggregation and analysis of volatile compound emissions for regional modelling, Atm. Env., 24, , Moody, T., J.D. Winkler, T. Wilson, S. Kersteter, The development and improvement of temporal allocation factor files, EPA report EPA-600/R , USEPA Office of Research and Development, Washington, D.C , Moran, M.D., M.T. Scholtz, C.F. Slama, A. Dorkalam, A. Taylor, N.S. Ting, D.Davies, P.A. Makar, and S. Venkatesh, An overview of CEPS1.0: Version 1.0 of the Canadian Emissions Processing System for regional-scale air quality models. Proc. 7 th AWMA Emission Inventory Symp., Oct , Research Triangle Park, North Carolina, Air & Waste Management Association, Pittsburgh Ryan, R., EPA Website [ accessed Nov 23,

51 EMISSIONS UNCERTAINTIES AND IMPACT ON REACTIVITY ASSESSMENTS A. Russell One of the great uncertainties in photochemical modeling of ozone is in the estimation of emissions. As reviewed in the recent NARSTO assessment, the uncertainties in the emissions, primarily of VOCs, is likely the greatest limitation in our ability to improve ozone simulations. This is true for both anthropogenic and biogenic emissions. This impacts our ability to quantify reactivity as well, but likely not to the same degree. Limited studies suggest that the impact of emissions inventory errors on reactivity quantification is significantly smaller than on the absolute ozone estimates. Examining, first, the issue of uncertainties in our ability to simulate ozone, Hanna et al., (1998), Gao et al., (1996), Russell and Dennis (1999) and Bergin et al., (1999) find that the uncertainties in absolute ozone are on the order of 25% (depending on study, all of which used different domains and parameters for uncertainty quantification). This can be compared to typical model performance which finds normalized errors on the order of 35% (Russell and Dennis, 1999). The reason for the increased error when looking at model performance is that the latter also includes errors in the meteorological fields (which were not completely treated in the other studies), grid resolution errors (which is linked to the commensurability problem), limitations in the model formulation and errors in the spatio-temporal distribution of emissions. We can not, yet, fully account for the myriad of contributors to uncertainty, so the above studies identified the major contributors for study. As an example of the importance of the emissions contributions to overall uncertainty, the Bergin et al., (1999) study found that emissions related uncertainties contributed between about 30 and 50% of the overall uncertainty in the ozone predictions. While errors in the emissions estimates may be a major contributor to the errors in the absolute ozone, it is not necessarily the case that they play the same role in quantifying VOC reactivity. The reason lies in expanding the reactivity estimation procedure as a Taylor series. Most of the terms between the base simulation and the perturbed case are the same, so they drop out when taking the difference. It is the second order terms that become important. A study by Yang et al., (1999) explicitly identified and the major contributors to overall uncertainty in relative reactivities of compounds in the Los Angeles area. While they found that uncertainties in the VOC emissions inventory (which were assumed to be log-normally distributed with a standard deviation of 2) accounted for about 8% of the total uncertainty ranking about 6th or 17 parameters tested), chemical parameters (e.g., the HCHO photolysis rate and the NO2 +HO reaction rate) had a greater influence. There overall reactivity estimates and uncertainties were similar to those found by Bergin et al., (1996), which did not include emissions uncertainties. Uncertainties in the NOx emissions had a significantly smaller impact. This study suggests that, at present, errors in the emissions estimates are not limiting our ability to quantify reactivity. However, this results of this study may not hold for other regions, and this finding should be tested. 6

52 References Bergin, M.S.; Russell, A.G.; and Milford, J.B. (1995). Environ. Sci. & Tech., 29(12): Bergin, M.S.; Russell, A.G.; and Milford, J.B. (1998). Environ. Sci. & Tech., 32(5): Bergin, M., Croes, B., Carter, W., Russell, A.G., and Seinfeld, J.H. (1998), Ozone Control and VOC Reactivity," E. Environmental Analysis and Remediation, Wiley-Interscience, pp Yang, YJ.; Khan, M.; Wilkinson, J.; and Russell, A.G. et al., (1999) Spatial uncertainty assessment of relative reactivities (in preparation) Russell, A.G. and Dennis, R. (1999) Photochemical Air Quality Modeling: NARSTO Critical Review, (submitted). Gao, D., Stockwell, W.R., and Milford, J.B.; J. Geophys. Res , 101, Hanna, S.R., Chang, J.C. and Fernau, M.E., Atmos. Env. 1998, 32(21)

53 STATUS OF USE OF ANTHROPOGENIC EMISSIONS IN AIR QUALITY ASSESSMENTS G. Sistla This report attempts to provide a brief summary on the estimation and use of anthropogenic emissions in the assessment of ozone air quality. The report draws heavily from the critical review papers developed under NARSTO 1-2, EPA 3-8 reports, and from Bergin et al (1998). While estimation and use of emissions in determining air quality has been a common practice, it is increasingly becoming important to characterize issues such as uncertainties and errors in the estimated emissions or in speciation characteristics, etc., and to assess their effect on air quality. Traditionally emissions are often estimated and reported on an annual basis in terms of their broad source categories stationary (point and area), mobile, and biogenic. However, in recent times there has been a need for information on emissions at a much more detailed level resulting in the development of emissions modeling systems The references 1 through 8 provide a wealth of information on the development of the inventories and some of the problems associated with them and their use in assessment of air quality. This report attempts to look at some of those issues from an end-user point of view and the need for improvement. Stationary Sources This broad categorization includes all sources that do not get classified to be associated under the mobile category. Thus for all practical purposes all stationary sources can be characterized as point sources, implying information on the location of the source, and pertinent stack information are available. However, it is often impossible to acquire relevant information for all but only those individual sources that are characterized by high emissions and associated with stacks, while the remaining sources are too numerous to be treated individually and therefore they fall under the area source category. Typical examples of the point sources are electric utility generators, while gasoline stations fall under the area source category. In addition, the distinction between point and area sources could be based upon a set emission level cut-off by pollutant. For example in the case of a severe ozone non-attainment area, the emission reporting requirements could seek information from sources that are at or above 5 tpy of VOCs or NOx, while in other areas of the state emissions reporting requirement are set at 100 tpy level. It is important to take into account the definition of point source and the attendant information that is available to characterize the source emissions since such information and quality of the data could vary significantly across geo-political sub-divisions of a state or a region. Further, it should be noted, that the quality of information sought to estimate emissions from point sources should not be limited only to the large emitters but allowance be made to acquire data for the low emitters as well, because such information provides critical backup in the development of controls as the cutoff limits are reduced. 8

54 As noticed in the draft EPA Guidance 4, the emissions data are to be speciated into chemical classes as required by the air quality models. Also, the emissions data are to be allocated on a diurnal basis or temporal allocation to provide hourly information for the air quality model. The speciation and temporal allocation processes are independent operations and in many instances currently utilize default profiles depending upon the nature and type of a source. Point and Area sources In many instances the emissions and other pertinent information that are gathered by the regulatory agencies is based upon actual operations or from the permit to operate the facility. In recent years the emissions have also been obtained from continuous emissions monitoring (CEMs) systems. These data require appropriate adjustment for use in air quality models. Such a manipulation is accomplished by emissions modeling tools such as EPS2 10, EMS95 11, SMOKES 12, FREDS 13, and CEPS 14. These tools have built in checks and balances to ensure that the data provided are within?standard? engineering practices in terms of information such as stack parameters, and emission estimates based upon the SCC. However, experience from the OTAG 15 has shown that there may be a need to?go-back?, as quite often the engineering data may be missing or faulty and even fundamental datum such as the location of the point source may be in error due to such things as data entry, conversion from English to Metric system of units or viceversa, etc. Although procedures used by states to generate emissions is undergoing changes through the EIIP 3-8 process, questions remain as to the allocation of resources and the cooperation of the facilities in providing these data requests. As noted above the temporal profile is often assumed in the apportionment of the emissions, and allocated to each hour. Along with the emissions, other engineering parameters are also needed in the air quality model. However, such information is often not available, resulting in the use of default data for the entire period of air quality simulation. Such a procedure may be reasonable when simulating episodic events that last 2 to 3 days and are generally associated with high 1-hr ozone levels that tend to occur on high temperature days and nearly stagnant conditions. However to assume or use default engineering data reflecting episodic conditions may not be meaningful and may result in over predictions in situations that cover extended periods that contain non-episode days. Examination of ambient air quality data suggests that exceedances of the 8-hr ozone NAAQS could occur under non-episodic conditions as well. Therefore there is a need to address on how to apportion the emissions between episodic and non-episodic days as well as how to estimate similarly the changes in the engineering parameters such as stack temperature and stack flow rate, etc. Also, it is important to address if there are changes or differences that may accompany in the level of VOC emissions and its composition between episodic and non-episodic conditions in the development of emissions. Another important parameter that may not be readily available is the information on the operational hours of a facility. Depending upon the time-of-the day and operational hours of the facility, emissions are to be apportioned on hour by hour basis and may have an effect on the air quality predictions if the emissions are wrongly characterized. 9

55 Often a majority of the area source emission categories are estimated based upon population as a surrogate parameter. The emission factors for area source categories are often adapted from EPA, unless data are available from local surveys or other such data bases. EIIP has recommended preferred and alternative methods for estimating for about 15 source categories that are predominantly VOC emitters. Often these methods require local surveys which will not be undertaken unless significant resources are directed by the state agencies for the development of emission inventories. Mobile sources Emissions from these sources are estimated in a top-down approach in which vehicle miles traveled (vmts) at the county level are utilized to estimate the emissions based on EPA s mobile emissions model. The emissions reported are in terms of NOx, CO, and VOCs, while the air quality model requires speciated data. Currently there are some indications that the states may wish to develop information from a bottom-up approach. Such an effort has been undertaken in the Lake Michigan Air Directors Consortium modeling work for the Greater Chicago area ozone air quality assessment. This requires collection of VMT and vehicle class information on a link basis to estimate emissions with the emissions model. Currently the mobile source model is undergoing updates and revisions. The model is not designed to provide speciated hydrocarbons but total hydrcarbons. The selection of an appropriate speciation profile that reflects the fuel characteristics of an area is a very important step in the emissions modeling system. For example, introduction of the reform-fuel (RFG) in ozone non-attainment areas in 1995 required the use of appropriate speciation profile and not the speciation profile of conventional gasoline. Often, such information may not be readily available, requiring the use of default profiles in air quality assessment. Speciation The emissions data are compiled and reported generally on an annual basis for the criteria pollutants NOx, CO, and VOCs at the county level. However the air quality models require the emissions data in terms of individual chemical species or class of species. This is often accomplished by using EPA?s models such as SPECIATE or FIRE, even though the majority of these data are based on testing of sources done in 1970's and early 1980's (see Ref. 1, 2). Use of updated speciation profiles should be considered a must for improving the emissions data base for use with air quality models. It should be noted that the VOCs estimated for some sources may not reflect the composition of the speciation profile due to elimination of some of those species through emission controls. Another important aspect of the speciated data is the need to ensure to what specific class of VOCs it is being referred to, such as - total organic gases (TOG), total hydrocarbons (THC), non-methane organic compounds (NMOC), non-methane organic gases (NMOG), reactive hydrocarbons (RHC), etc. Also, state level controls may result in revisions to manufacturing processes and chemical formulation that may be reflective of that state only and should be accounted for in the preparation of inventories for air quality applications. 10

56 Future work It is recommended that the EIIP procedures be followed to the maximum extent where feasible in the development of emissions inventory. The data collected should include besides operational information, stack sampling or other information to facilitate speciation of the VOCs. Use of default information to estimate the emissions should be discouraged as there may be no incentive to improve the data. Industry participation in the development of speciation profiles is a critical first step towards addressing issues and controls based upon HC reactivity. References 1. NARSTO Critical Review Paper: Emissions from Stationary Sources, (1998), M. Placet, C. Mann, R. Gilbert, and M. Niefer 2. NARSTO Critical Review Paper:Mobile Sources Critical Review, (1998), R. Sawyer, R. Harley, S. Cadle, J. Norbeck, R. Slott, and H. Bravo 3. Procedures for the Preparation of Emission Inventories for Carbon Monoxide and Precursors of Ozone, Volume I, (1991), USEPA, EPA-450/ Emission Inventory Guidance for Implementation of Ozone and Particulate Matter National Ambient Air Quality Standards (NAAQS) and Regional Haze Regulations, (1998) Draft Report 5. Introduction to the Emission Inventory Improvement Program, Volume 1, (1997), USEPA 6. Preferred and Alternative Methods for Estimating Air Emissions, Volume II, (1996), Point Sources Preferred Methods, EIIP, USEPA 7. Preferred and Alternative Methods for Estimating Air Emissions, Volume III, (1996), Area Sources Preferred Methods, EIIP, USEPA 8. Preferred and Alternative Methods for Estimating Air Emissions, Volume IV, (1996), Mobile Sources, EIIP, USEPA 9. Ozone Control and VOC Reactivity, (1998), M. Bergen, A. Russell, W. Carter, B. Croes, and J. Seinfeld in Encyclopedia of Environmental Analysis and Remediation, Edited by R. Meyers, John Wiley & Sons, Inc. 10. User s Guide for the Urban Airshed Model Volume IV: User s Manual for the Emissions Preprocessor System 2.0, (1992), EPA-450/ D(R) 11. Technical Formulation Document: SARAMAP/LMOS Emissions Modeling System (EMS- 95), (1994), AG-90/TS26 and AG-90/TS27, prepared for the Lake Michigan Air Directors Consortium, Des Plaines, IL and The Valley Air Pollution Study Agency, Technical Support Division, Sacramento, CA 11

57 12. Sparse Matrix Operator Kernel Emissions (SMOKE) Modeling System (1996), iceis.mcnc.org/edss/emissions/index.html, MCNC Environmental Programs, Research Triangle Park, NC 13. Flexible Regional Emissions Data System (FREDS) Documentation for the 1985 NAPAP Emissions Inventory, (1989), EPA-600/ Moran, M.D., M.T. Scholtz, C.F. Slama, A. Dorkalam, A. Taylor, N.S. Ting, D.Davies, P.A. Makar, and S. Venkatesh, An overview of CEPS1.0: Version 1.0 of the Canadian Emissions Processing System for regional-scale air quality models. Proc. 7 th AWMA Emission Inventory Symp., Oct , Research Triangle Park, North Carolina, Air & Waste Management Association, Pittsburgh. 15. OTAG Final Report, (1998) at 12

58 BIOGENIC EMISSIONS AND CHEMICAL INTERACTIONS P. A. Makar The current state of the science for measurements and modelling of biogenic emissions has been extensively reviewed in Guenther et al. (1998). Some of the main conclusions of this work with regards to reactivity will be repeated here, as well as chemical effects from recent work not considered in the above reference. Several sources referenced in Guenther et al. (1998) have shown that model predictions of ozone mixing ratios are strongly influenced by the effect of biogenic VOC emissions, and can determine the choice of a NOx or VOC control strategy. Accurate estimates of biogenic emissions are therefore of great importance for estimating atmospheric reactivity. The factors for determining biogenic emissions include area-average factors (emission per unit area of ground for standard meteorological conditions), emission activity factors (canopymicroclimate; leaf temperature and photosynthetically active radiation being dominant effects) and escape efficiency factors (the extent to which deposition decreases the actual emission to the above canopy air) (Guenther et al., 1998). In addition, recent work by Makar et al. (1998) and Forkel et al (1998) suggests that chemical losses may result in underestimates of biogenic emissions. The uncertainties associated with these factors and effects are a useful means of setting priorities for improving knowledge of biogenic emissions. Uncertainties for the area average factors for individual tree species (which dominate biogenic emissions) were estimated at 20 to 50%, while crop emissions and non-terpene emissions from all plant species were estimated to have uncertainties of a factor of 3 or more. Genetic variability within and between plant species have shown large variations in emissions (factor of 4 to 6 within same species; Guenther et al. (1998)). In addition, the construction of vegetation databases was discussed in the context of the BEIS3 model. Uncertainties associated with composition and foliar density of landcover were given as ranging from 25% to a factor of 3, depending on the landscape. Some of the basic data for constructing the database was unavailable in Canada, necessitating the use of approximations for forest density parameters. The reduction of these uncertainties through the continuation of ongoing measurement programs was identified as a research priority. The uncertainties associated with light and temperature emission activity factors were estimated at 15 to 25%, with seasonal factors being greater than 25%. Activity factors associated with wounding (damage of leaves by environmental conditions) had uncertainties much greater than 25%. Stomatal conductance factors, which affects the emission rate of oxygenated compounds, were assumed to have uncertainties of 25 to 50%. Leaf temperature parameterizations and those for soil temperature were assumed to have maximum uncertainties of 30 and 25%, respectively. 13

59 Escape efficiency was assumed to have an uncertainty of 5 to 30%. One effect not considered for emissions estimates which has appeared in two recent studies (Makar et al., 1998; Forkel et al, 1998), is the effect of photochemical reactivity on estimates of emissions fluxes. Both of these modelling studies utilized one-dimensional canopy models to simulate biogenic emissions. In each case, detailed photochemical models were coupled with emission factor formulae and vertical mixing parameterizations. The effect of chemical removal on emissions estimates was examined by running chemistry and nochemistry simulations. In Makar et al. (1998), neglecting effects of chemical losses was found to result in an underestimate of isoprene emission fluxes by as much as 40%. The study of Forkel et al. (1998) found a 20% isoprene emission underestimate if chemistry was neglected. Makar et al. (1998) noted that chemical loss effects would only be significant when the emitted compound was the most reactive in the ambient atmosphere; this has been confirmed by more recent work by Forkel (personal communication, 1998) where the absence of chemical losses was found to have a greater impact on the estimates of emissions of more reactive terpenes than for isoprene. From the above uncertainties, the improvement of area average factors and land-use categories have the greatest need for further research, with activity factors, escape factors and chemical losses having a secondary and approximately equal effect on the overall uncertainty in the emissions. The effects of plant wounding could be considerable; research is required to determine the best means of estimating these effects on emissions. Other reactivity related conclusions from Guenther et al (1998) included noting that pre- BEIS3 emissions models greatly overestimate alkane and aromatic compounds, while noting that alkene emissions do occur, model validation is required for conditions outside of midday, summer conditions, emissions of C 1 - C 3 species and oxygenated VOCs need to be better characterized, and expected changes in species composition and biomass density require better characterization. References Guenther, A., C. Geron, T. Pierce, B. Lamb, P. Harley and R. Fall, Natural emissions of nonmethane volatile organic compounds, carbon monoxide, and oxides of nitrogen from North America, NARSTO critical review paper posted on NARSTO website [ accessed Nov. 23, Forkel, R., W.R. Stockwell and R. Steinbrecher, A multilayer canopy/chemistry model to simulate the effect of in-canopy processes on the emission rates of biogenic VOC, Proceedings, APMS 98 - International Conference on Air Pollution Modelling and Simulation, Ecole Nationale des Ponts et Chaussees/Institut National de Recherche en Informatique et en Automatique, Champs-sur-Marne, France, , Makar, P.A., J.D. Fuentes, D. Wang, R.M. Staebler and H.A. Wiebe, Chemical processing of biogenic hydrocarbons within and above a temperate deciduous forest, J. Geophys. Res., (in press),

60 ENVIRONMENTAL CONDITIONS E. P. Olaguer 1, D. W. Byun 2, J. A. Dege 3, and P. J. Ostrowski 4 1 Dow Chemical Co. 2 NOAA 3 Dupont Co. 4 Occidental Chemical Co. CONTENTS CONTENTS... 1 ENVIRONMENTAL FACTORS AFFECTING REACTIVITY... 1 ATMOSPHERIC STRUCTURE AND DYNAMICS... 2 POTENTIAL IMPACTS OF LONG-RANGE TRANSPORT MECHANISMS... 4 VARIABILITY OF CURRENT REACTIVITY SCALES... 6 SUMMARY AND RECOMMENDATIONS... 7 REFERENCES... 8 ENVIRONMENTAL FACTORS AFFECTING REACTIVITY Reactivity is, strictly speaking, not a constant, inherent property of a chemical species, since ambient conditions have a substantial influence on the amount of ozone that can be produced by a VOC within a specific airshed. The photolysis of atmospheric constituents, for example, depends on the overhead ozone column, the distribution of clouds, and the solar zenith angle. Chemical formation of ozone likewise depends on ambient pressure, temperature, humidity, the absolute VOC and NOx concentrations, the VOC/NOx ratio, and the overall supply of radicals (especially hydroxyl, which determines the atmospheric lifetimes of the majority of VOCs). These factors are in turn influenced by atmospheric transport to and from an airshed by prevailing winds, by horizontal and vertical mixing within the airshed, and by wet and dry removal processes. Anthropogenic and biogenic emissions of VOCs and NOx themselves depend on wind, humidity, atmospheric stability, and temperature. Meteorology is therefore an important aspect of any discussion of the concept of reactivity. The familiar example of the Los Angeles basin, wherein so-called inversion layers of stable, warm air combine with the effects of local topography to produce frequent smog episodes, illustrates how critical the peculiar meteorological conditions of an airshed can be in accounting for ozone formation. The central problem involving the impact of environmental factors on reactivity is whether or not a relative reactivity scale can be defined which is only weakly affected by

61 variations in environmental conditions within and among different airsheds, so that it can be applied on a national basis and with a reasonable expectation of effectiveness. For example, will the relative reactivity of VOCs within the Los Angeles basin differ substantially from that determined within the northeast corridor of the United States, where long range transport of ozone and precursors, as opposed to local trapping of air pollution, is an important phenomenon? A related issue is whether or not the scientific methodologies that have hitherto been used to examine this question are themselves capable of yielding an unequivocal answer. The resolution of these difficulties depends partly on how the problem of ozone control is defined. Given that a change in the federal ozone standard (now 80 ppb averaged over 8 hours) has occurred, effectively giving more attention to chronic ozone exposure over large regions than to episodal peaks within urban airsheds, should reactivity be defined over multiple days and large regions rather than over single day urban episodes, as has previously been the case? If the answer to this last question is yes, then are the vital processes which may determine large-scale reactivity captured by current scientific models? The aim of this chapter is to enumerate and describe some of these vital processes so that models for environmental conditions can be appropriately evaluated, and the corresponding implications for future reactivity research identified. ATMOSPHERIC STRUCTURE AND DYNAMICS Before discussing the merits and demerits of any particular approach to reactivity, it is necessary to first understand the structural and dynamical features of the atmosphere that are vital to tropospheric ozone formation and transport. This includes the vertical stratification, stratospheric air intrusions, the boundary layer, atmospheric convection, the existence of westerly jet streams in the free troposphere, the role of large-scale weather (i.e., synoptic systems), microscale influences, and coastal meteorology. The troposphere is defined as the region of the atmosphere immediately above the surface where temperature generally decreases with height. The top of the troposphere ranges from about 8 km at high latitudes to a maximum of about 16 km at the equator. The region of the atmosphere immediately above the troposphere and where the temperature generally increases with height is known as the stratosphere. The stratosphere is typically rich in ozone because of the equatorial maximum in photochemical ozone production at an altitude of about 35 km, and because of poleward and downward transport of ozone from this region. Occasionally, tropopause folding associated with certain weather events can result in stratospheric air intrusions into the troposphere (one aspect of so-called stratospheric-tropospheric exchange ), during which downward transport of ozone-rich air produces sudden peaks in ground-level ozone (Danielsen, 1968; Ebel et al., 1991). This fact alone indicates the potential importance of downward subsidence throughout the depth of the troposphere in determining environmental conditions near the surface. 2

62 The troposphere itself is generally divided into the boundary layer, in which the frictional drag of the surface is important and where rapid vertical mixing occurs, and the free troposphere, in which the prevailing winds away from equatorial regions are determined largely by a balance between horizontal pressure gradients and the Coriolis force due to the earth s rotation (the socalled quasi-geostrophic balance ). The height of the boundary layer increases with convective instability due to vertical temperature gradients, typically undergoing a diurnal variation and ranging from as low as 100 meters at night to as high as 3 kilometers during the day. This diurnal pattern can affect the VOC/NOx ratio and radical density near the surface through changes in the effective diluting volume and through encroachment of free tropospheric air into the boundary layer. Summertime conditions associated with the development of thunderstorms and cumulus towers can lead to a phenomenon known as penetrating or deep convection, wherein surface pollutants are rapidly (within an hour) lofted from the boundary layer into the free troposphere by intense convective plumes, and pollutants in the free troposphere are entrained into the boundary layer by a corresponding slow subsidence (within about 50 hours) in large regions around the convective plumes (Lelieveld and Crutzen, 1994). Thompson et al. (1994) showed that the central United States acts as a convective chimney for the country by venting large amounts of carbon monoxide and other ozone precursors from the boundary layer to the free troposphere, possibly contributing to high background levels of ozone in the eastern United States. Once in the free troposphere, pollutants undergo faster eastward dispersion than they would otherwise near the surface, due to upper level winds associated with the climatological westerly jet stream centered at approximately 30 degrees north. The pollutants can then reenter the boundary layer either through convectively-induced subsidence and subsequent encroachment into the mixed layer, or within regions of subsiding air associated with traveling weather or synoptic systems. Carmichael et al. (1998) demonstrated using simulations of a 3- D Eulerian model and comparisons to observations, that strong downward fluxes of ozone from the upper troposphere associated with the passage of cold fronts can play a critical role compared to chemistry in accounting for variations in ozone near the surface. Cooper et al. (1998) likewise found evidence of strong ozone subsidence behind cold fronts from observations taken during the Atmosphere/Ocean Chemistry Experiment (AEROCE), together with indications that such postfrontal subsidence can bring substantial amounts of stratospheric ozone into the troposphere. Within an urban airshed, environmental conditions are strongly affected by variations in the local heat balance, in air flow over topography, in source emissions, and in wet and dry deposition. During stagnant conditions, fast nocturnal winds can rapidly transport pollutants out of urban areas to rural regions (Banta et al., 1998; Meagher et al., 1998), whereas land-sea breeze circulations in coastal areas can re-circulate stagnant air back into an urban airshed, thus prolonging pollution episodes beyond a day and increasing the reactivities of slow-reacting compounds (McNair et al., 1994). It is also possible that in coastal cities such as Houston and Los Angeles, the presence of chlorine atoms from sea salt may account for more than 10% of initial VOC oxidation, thus significantly impacting ozone formation (Livingston and Finlayson- Pitts, 1991). 3

63 POTENTIAL IMPACTS OF LONG-RANGE TRANSPORT MECHANISMS The coupling of the boundary layer and the free troposphere has important implications for long range reactivity because the lifetimes of some important ozone precursors such as PAN increase substantially in the colder, less humid environment of the free troposphere. Ozone itself lives considerably longer when not subject to surface deposition. Any transport mechanism, therefore, that lofts ozone or PAN into the free troposphere and then subsequently entrains it into the boundary layer may have an important chemical influence on surface air at large distances downwind of the jet stream. In addition, longer-lived organics such as methane and carbon monoxide can have a significant impact when transported downwind. Jeffries (1993) has shown that the majority of ozone formation in a typical urban airshed may actually be attributable to such longer-lived compounds, even though they are considerably less reactive than most VOCs when gauged by existing urban reactivity indices. McKeen et al. (1991) likewise showed using a 3-D regional model of the eastern United States that methane and carbon monoxide contribute significantly to regionally averaged net ozone formation. Given that air upwind of an urban area can have an ozone level already close to the new federal ozone standard (Valente et al., 1998), any ozone attributable to long-range transport will be of considerable importance. A recent study by Berkowitz et al. (1998) examined airborne chemistry and meteorological measurements over the eastern seaboard of the U.S. using 4-D data assimilation within a mesoscale model, together with a Lagrangian particle dispersion model. They concluded that for a particular scenario on August 31, 1995, an elevated layer of ozone, PAN, and other ozone precursors just above the boundary layer was brought to the surface with the onset of convection. They further suggested that such elevated layers of pollutant frequently form over the northeastern United States due to lofting of pollutants into the free troposphere from distant sources, and that a significant portion of ozone at a particular location may be due to the entrainment of such layers into the convective boundary layer. When ozone is entrained into the boundary layer, there is more formation of hydroxyl radicals, which then increases the local reactivity of VOCs. When PAN or other organic nitrates are entrained into the boundary layer, they act as potential sources, not only of radicals but also of NOx, so that VOCs which form nitrates can act as remote sources of NOx downwind. This effective remote supply of NOx and radicals can therefore increase the long-range reactivity of parent VOCs, especially when the associated NOx reservoirs are delivered to low-nox environments, where the efficiency of ozone formation is considerably greater than in urban areas (Lin et al., 1988), and where there is considerably less competition from in-situ NOx sources. In many areas of the country, local biogenic emissions of compounds such as isoprene may compete with or even dominate local anthropogenic sources of VOC, and some studies (e.g., Cardelino and Chameides, 1995) have used urban airshed models to deduce areas where local VOC control may be ineffective compared to NOx control for such reasons. However, a mechanism by which VOCs contribute to the NOx budget far downwind may conceivably allow remote urban VOCs which combine with intense mobile and stationary NOx emissions to 4

64 compete effectively with in situ biogenic sources of VOCs as regional ozone precursors. Of course, biogenic emissions can themselves form transportable organic nitrates in addition to releasing radicals that contribute to local ozone formation. The wet and dry deposition rates of biogenic organic nitrates, however, may be such that they are actually an effective route for removal of NOx from polluted air masses (O Brien et al., 1995). Measurements indicate that, even in remote regions of the troposphere, PAN and other organic nitrates may be contributing significantly to the total budget of reactive nitrogen (Atlas et al., 1992; Singh et al., 1992). A proper assessment of the role of organic nitrates in regional and global ozone pollution, however, will require a major improvement in both the analytical detection of organic nitrates and the accounting of organic nitrate degradation products within chemical mechanisms, including those that partition significantly in the aerosol phase. Duncan and Chameides (1998) have recently examined the potential effects of urban control strategies on exports of ozone and ozone precursors from urban airsheds to the largerscale troposphere using a simple urban plume model. They discovered that while NOx reductions were always more effective than VOC reductions in controlling the direct export of ozone from urban areas, the relative importance of VOC and NOx reductions in controlling the export of ozone precursors from the urban atmosphere was more complicated to describe. In particular, while NOx reductions always reduced the exports of nitrogen oxide species and acetone, they enhanced the export of paraffins and significantly changed the speciation of the exported reactive nitrogen. Moreover, while VOC reductions decreased the export of PAN and other organic nitrates, they could also increase the export of NOx and nitric acid. Duncan and Chameides (1998) concluded that a proper assessment of the impact of changing export rates of urban ozone precursors on tropospheric ozone can only be carried out with large-scale models. The application of new-generation 3-D regional models to the assessment of reactivity becomes especially critical if long-range transport mechanisms involving boundary layer-free troposphere coupling are found to contribute substantially to the NOx and ozone budgets in areas that are out of compliance with the new federal ozone standard. It will be important to determine the boundary conditions for such models accurately using hemispheric and global models, especially if hemisphere-wide transport significantly impacts regional background concentrations of ozone, nitrates, and long-lived organics. That this may in fact be the case is suggested by Jacob et al. (1993), who employed a 3-D regional model to study the effect of a 50% reduction in U.S. NOx emissions. They found that this substantial emissions reduction led to a weak decrease in rural ozone over the eastern U.S. of only 15% due to the substantial import of ozone from outside North America. Moreover, they found that the total U.S. export of ozone and precursors to be about one-third that due to stratospheric-tropospheric exchange. The potential impact of such exports across the Atlantic is made clear in a study by Builtjes (1992), who concluded from experiments with the LOTOS 3-D model that persistently high regional ozone in Europe is affected by transport over the entire Northern Hemisphere. 5

65 VARIABILITY OF CURRENT REACTIVITY SCALES Assessment of the urban reactivity scales defined by Carter (1994) are discussed in detail in a later chapter. We mention very briefly, however, some pertinent aspects of these assessments in order to point out the role that models play in examining the potential influence of environmental variability on reactivity scales. The current use of reactivity within the United States is an outgrowth of attempts to solve the urban ozone problem as it was defined and understood during the 1970s through the major part of this decade, when controlling episodal peaks in ozone above 120 ppb near urban cores was the primary goal of air quality strategies. Carter s reactivity scaling approach acknowledges the role of environmental variability within this framework by averaging reactivities calculated with a simple box/trajectory model over 39 U.S. cities. Among the 39 scenarios, it is possible for a compound to go from being fairly reactive to having a negative reactivity (e.g., toluene), although the basic trend in the reactivity of most organics is maintained (i.e., highly/poorly reactive compounds under one scenario generally remain highly/poorly reactive under another). The impact of environmental variability is apparently lessened when relative rather than absolute reactivities are used, and when high NOx conditions are assumed within each scenario (Carter, 1994). The assumption of high NOx conditions, however, does not necessarily correspond to conditions under which exceedances of the new federal ozone standard may occur, particularly outside urban areas. The issue of environmental variability has been further explored in comparisons between reactivities computed using the box model approach and those computed using 3-D Eulerian models, which cover domains with a wide range of environmental conditions, from NOx-rich urban centers to VOC-rich areas downwind. The studies of Russell et al. (1995) and Bergin et al. (1995) showed that, with some notable exceptions such as aromatics and photoreactive species, multi-day reactivities derived from the CIT 3-D urban airshed model were in good agreement with Carter s box model-derived single-day reactivities in the Los Angeles basin. The CIT model, however, shares a common deficiency with Carter s box model, in that it is largely confined to the boundary layer. This feature eliminates the transport mechanisms involving the coupling of the boundary layer to the free troposphere referred to in previous sections. A similar criticism applies to the Photochemical Ozone Creation Potential (POCP) scale of Derwent and Jenkin (1991), which like the reactivity scales of Carter, is computed using a box/trajectory model, but for longer trajectories over multiple days. It is therefore not yet known whether the apparently good correlation between box/trajectory model reactivities and more sophisticated 3- D Eulerian model reactivities is retained when the vertical and horizontal domains of the 3-D model are expanded and the appropriate transport mechanisms considered, particularly for areas such as the Ozone Transport Assessment Group (OTAG) region, where long-range transport is a key factor. 6

66 SUMMARY AND RECOMMENDATIONS Current urban reactivity scales presume that air near the surface moves slowly and is confined to the boundary layer. They do not consider long range transport mechanisms involving the coupling of the boundary layer to the free troposphere, wherein ozone and its precursors are rapidly vented aloft and quickly transported downwind of the mid-latitude jet stream, where they can be re-entrained into the boundary layer. If reactivity scales are to be relevant to the new ozone standard of 80 ppb averaged over 8 hours, then they must account for such transport mechanisms for the following reasons: Long-range transport of ozone and ozone precursors via the free troposphere has the potential to influence the level of radicals in remote urban and rural areas, thereby influencing the local reactivities of VOCs in those areas. Long-range transport of ozone precursors via the free troposphere has the potential to influence the NOx budget in remote rural areas, through reservoir species such as PAN and other organic nitrates that do not easily rain out of the atmosphere. Long-range transport of ozone precursors via the free troposphere has the potential to produce significant quantities of ozone in remote urban and rural areas when compared to the difference between the new federal ozone standard (80 ppb) and the level of background tropospheric ozone (30 ppb). Regional, 3-D Eulerian models that adequately resolve both the boundary layer and the free troposphere are the only tools that can properly account for long-range transport mechanisms involving boundary layer-free troposphere coupling. We therefore recommend that their use in reactivity research be expanded to properly design regulatory controls based on reactivity and to assess their effectiveness in achieving and maintaining compliance with the new federal ozone standard on a national basis. We further recommend research in the following areas to better address the issues raised in this chapter: The role of upper and lateral boundary conditions for ozone and ozone precursors in determining the reactivities of VOCs in urban and rural airsheds. Improved representations of convective, mesoscale, and synoptic-scale transport in largescale reactivity models, including deep convection, frontal motions, and tropopause folding events. More complete representation of radical sources in reactivity models, including chlorine radicals from sea salt in coastal airsheds. Better resolution of organic nitrates in chemical mechanisms, including both soluble multi-functional organic nitrates that may deplete NOx from polluted air masses and nonsoluble organic nitrates that can deliver NOx to remote rural environments. 7

67 Rigorous measurements of the total reactive nitrogen budget in urban and rural areas within the United States to provide more extensive ground truth for both chemical mechanisms and airshed models. More emphasis on evaluating export-oriented approaches to reactivity. Large-scale 3-D modeling assessments of the effectiveness of existing and alternative reactivity-based control strategies. REFERENCES Atlas, E.L., B.A. Ridley, G. Hubler, J.G. Walega, M.A. Carroll, D.D. Montzka, B.J. Huebert, R.B. Norton, F.E. Grahek, and S. Schauffler, 1992: Partitioning and budget of NOy species during the Mauna Loa Observatory Photochemistry Experiment, J. Geophys. Res., 97, 10,449. Banta, R.M., C.J. Senff, A.B. White, M. Trainer, R.T. McNider, R.J. Valente, S.D. Mayor, R.J. Alvarez, R.M. Hardesty, D. Parrish, and F.C. Fehsenfeld, 1998: Daytime buildup and nighttime transport of urban ozone in the boundary layer during a stagnation episode, J. Geophys. Res., 103, 22,519. Bergin, M.S., A.G. Russell, and J.B. Milford, 1995: Quantification of individual VOC reactivity using a chemically detailed, three-dimensional photochemical model, Environ. Sci., Technol., 29, Berkowitz, C.M., J.D. Fast, S.R. Springston, R.J. Larsen, C.W. Spicer, P.V. Doskey, J.M. Hubbe, and R. Plastridge, 1998: Formation mechanisms and chemical characteristics of elevated photochemical layers over the northeast United States, J. Geophys. Res., 103, 10,631. Builtjes, P.J.H., 1992: The LOTOS- Long Term Ozone Simulation- project: summary report, TNO Environmental and Energy Research, Report No. IMW-R 92/240, Delft, the Netherlands. Cardelino, C.A. and W.L. Chameides, 1995: An observation-based model for analyzing ozone precursor relationships in the urban atmosphere, Air and Waste, 45, 161. Carmichael, G.R., I. Uno, M.J. Phadnis, Y. Zhang, and Y. Sunwoo, 1998: Tropospheric ozone production and transport in the springtime in east Asia, J. Geophys. Res., 103, 10,649. Carter, W.P.L., 1994: Development of ozone reactivity scales for volatile organic compounds, Air and Waste, 44,

68 Cooper, O.R., J.L. Moody, J.C. Davenport, S.J. Oltmans, B.J. Johnson, X. Chen, P.B. Shepson, and J.T. Merrill, 1998: Influence of springtime weather systems on vertical ozone distributions over three North American sites, J. Geophys. Res., 103, 22,001. Danielsen, E.F., 1968: Stratospheric-tropospheric exchange based on radioactivity, ozone and potential vorticity, J. Atmos. Sci., 25, 502. Derwent, R.G. and M.E. Jenkin, 1991: Hydrocarbons and the long-range transport of ozone and PAN across Europe, Atmos. Env., 25A, Duncan, B.N. and W.L. Chameides, 1998: Effects of urban emission control strategies on the export of ozone and ozone precursors from the urban atmosphere to the troposphere, J. Geophys. Res., 103, 28,159. Ebel, A., H. Hass, H.J. Jakobs, M. Laube, M. Memmesheimer, and A. Oberreuter, 1991: Simulation of ozone intrusion caused by a tropopause fold and cut-off low, Atmos. Env., 25A, Jacob, D.J., J.A. Logan, G.M. Gardner, R.M. Yevich, C.M. Spivakovsky, and S.C. Wofsy, 1993: Factors regulating ozone over the United States and its export to the global atmosphere, J. Geophys. Res., 98, 14,817. Jeffries, H., 1993: A different view of ozone formation: fuel reformulation alone may not reduce urban ozone, Fuel Reformulation, 3, 58. Lelieveld, J. and P.J. Crutzen, 1994: Role of deep cloud convection in the ozone budget of the troposphere, Science, 264, Lin, X., M. Trainer, and S.C. Liu, 1988: On the nonlinearity of the tropospheric ozone production, J. Geophys. Res., 93, 15,879. Livingston, F.E. and B.J. Finlayson-Pitts, 1991: The reaction of gaseous N 2 O 5 with solid NaCl at 298-K- Estimated lower limit to the reaction probability and its potential role in tropospheric and stratospheric chemistry, Geophys. Res. Lett., 18, 17. McKeen, S.A., E.-Y. Hsie, and S.C. Liu, 1991: A study of the dependence of rural ozone on ozone precursors in the eastern United States, J. Geophys. Res., 96, 15,377. McNair, L.A., A.G. Russell, and M.T. Odman, 1994: Airshed model evaluation of reactivity adjustment factors calculated with the Maximum Incremental Reactivity scale for transitional-low emission vehicles, Air and Waste, 44, 900. Meagher, J.F., E.B. Cowling, F.C. Fehsenfeld, W.J. Parkhurst, 1998: Ozone formation and transport in the southeastern United States: Overview of the SOS Nashville/Middle Tennessee ozone study, J. Geophys. Res., 103, 22,213. 9

69 O Brien, J.M., P.B. Shepson, K. Muthuramu, C. Hao, H. Niki, D.R. Hastie, R. Taylor, and P.B. Roussel, 1995: Measurements of alkyl and multifunctional organic nitrates at a rural site in Ontario, J. Geophys. Res., 100, 22,795. Russell, A., J. Milford, M.S. Bergin, S. McBride, L. McNair, Y. Yang, W.R. Stockwell, and B. Croes, 1995: Urban ozone control and atmospheric reactivity of organic gases, Science, 269, 491. Singh, H.B., D. O Hara, D. Herlth, J.D. Bradshaw, S.T. Sandholm, G.L. Gregory, G.W. Sachse, D.R. Blake, P.J. Crutzen, and M.A. Kanakidou, 1992: Atmospheric measurements of peroxyacetyl nitrate and other organic nitrates at high latitudes: possible sources and sinks, J. Geophys. Res., 97, 16,511. Thompson, A.M., K.E. Pickering, R.R. Dickerson, W.G. Ellis, Jr., D.J. Jacob, J.R. Scala, W.-K. Tao, D.P. McNamara, and J. Simpson, 1994: Convective transport over the central United States and its role in regional CO and ozone budgets, J. Geophys. Res., 99, 18,703. Valente, R.J., R.E. Imhoff, R.L. Tanner, J.F. Meagher, P.H. Daum, R.M. Hardesty, R.M. Banta, R.J. Alvarez, R.T. McNider, and N.V. Gillani, 1998: Ozone production during an urban air stagnation episode over Nashville, Tennessee, J. Geophys. Res., 103, 22,

70 VOC REACTIVITY SCIENCE ASSESSMENT: EXECUTIVE SUMMARY 5/5/99 CONTENTS BACKGROUND...1 ASSESSMENT OF SCIENTIFIC ISSUES...3 Atmospheric Chemistry...3 Aerosol Forming Potential...7 Emissions Data...9 Volatility and Fate...11 Air Quality Models...12 Environmental Conditions...14 Reactivity Assessments...15 Persistent Organic Pollutants...17 SUMMARY OF RECOMMENDATIONS...18 Atmospheric Chemistry...18 Laboratory Studies...18 Computational and Modeling Techniques...19 Smog Chamber Studies...20 Aerosol Formation Potential...20 Emissions Data...21 Volatility and Fate...21 Air Quality Models...22 Environmental Conditions...22 Reactivity Assessments...23 Persistent Organic Pollutants...24 REFERENCES...25 LIST OF CONTRIBUTORS...28 BACKGROUND When volatile organic compounds (VOCs) are emitted into the atmosphere, they can undergo photochemical reactions that may contribute to the formation of ground-level ozone, which exceeds air quality standards in many areas. For that reason, emissions of VOCs have been subject to controls for a number of years, and significant additional controls are believed to be necessary before air quality standards can be met. Certain VOCs may also contribute to the formation of atmospheric particulate matter (PM), form toxic oxidation products, or have other impacts on the environment. The emissions of certain VOCs are regulated or banned because 1

71 they are highly toxic or deplete the stratospheric ozone layer. However, other than this, the need to reduce ozone has been the main factor driving the regulation of VOC emissions in the United States. It has been realized for a number of years that not all VOCs are equal in their effects on ground-level ozone formation. Some VOCs react so slowly they have almost no effect on ozone pollution episodes, others not only form ozone themselves but also enhance ozone formation from other VOCs, and others actually inhibit ozone formation. In recognition of this, the U.S. EPA has had a policy of exempting from VOC regulations compounds that react extremely slowly or, more recently, compounds which have been shown to be ozone inhibitors. However, the vast majority of VOCs have been regulated as if they all had the same effect on ozone, even though this is known not to be the case. The impact of a VOC on formation of ozone or other measures of air quality is referred to as its atmospheric reactivity. VOC regulations that take into account differences in VOC reactivity have the potential of being much more cost-effective than present policy, and would eliminate the need for the arbitrary dividing line in the current exemption policy. However, there are significant uncertainties on how VOC reactivities should be quantified and determined, and there are major unresolved policy issues that affect what scientific research is most needed. In view of this, the EPA and other regulatory agencies joined with industry groups and interested researchers to form the Reactivity Research Working Group (RRWG), to coordinate policyrelevant research related to VOC reactivity. The RRWG mission statement is as follows: Our mission is to provide an improved scientific basis for reactivity-related regulatory policies. That will be accomplished by bringing together all parties actively interested in sponsoring, planning, performing or assessing policyrelevant scientific research on the reactivities of organic compounds emitted to ambient air, as related to the formation of ozone, PM2.5, and regional haze. This is for the purposes of coordinating such research and defining potential applications, while continuously involving key policy makers. One of the first tasks of the RRWG is to produce assessments of policy and scientific issues related to reactivity. This document gives the result of the assessment of the scientific issues. Its objective of this assessment is to summarize the current state of the science related to quantifying VOC impacts in the atmosphere, and to summarize the critical research areas most needed to reduce the uncertainties involved. Contributors include many of the scientists who have conducting research in these areas, as well as other interested participants in the RRWG. This document, together with the RRWG policy issues document, will then be used as a basis for formulating a research plan to provide an improved scientific basis for reactivity-related regulatory policies. 2

72 ASSESSMENT OF SCIENTIFIC ISSUES There are many factors that affect the impact of a VOC on the atmosphere. These include how much is emitted into the gas phase, how rapidly it reacts, how its reactions affect ozone and other impacts, how environmental conditions affect how rapidly it reacts and the impacts of its reactions, and the volatilities, reactivities and impacts of its oxidation products. Because it is difficult to duplicate in the laboratory all the environmental factors that affect VOC reactivity, the only practical way to quantify a VOC s impact in the atmosphere is to calculate it using computer airshed models. This requires a mechanism for the VOC s relevant atmospheric reactions, and a model for the environment where the VOC is reacting. If either is incorrect, then the model predictions for the VOC s impact in that environment will also be incorrect, at least to some extent. Furthermore, since the VOC s impact depends on the environment, the choice of the environment to carry out the assessment will affect the results. Since environmental conditions are highly variable, knowledge of the distribution of environmental conditions is important for determining the most appropriate reactivity scale to use in regulations affecting large regions or the entire United States. Clearly, there are many scientific issues and research areas involved in the quantification of VOC reactivity. For the purpose of this assessment, these have been divided into seven different areas, which are discussed in separate chapters of this assessment document. These are briefly summarized below. Atmospheric Chemistry Model predictions of a VOC s impact can be no more reliable than the chemical mechanism used in the model to represent its atmospheric reactions. Although the rate constants for initial reaction consuming most VOCs in the atmosphere are known or can be estimated, the subsequent reactions for most VOCs are complex and incompletely understood. Furthermore, many of the oxidation products formed are unknown or uncertain. Nevertheless, significant progress has been made in recent years in improving our knowledge of and ability to predictively model the ozone formation potentials of many VOCs. However, ozone predictions for many other VOCs are still highly uncertain, and predictions of other impacts besides ozone require knowledge of much more of the details of the mechanisms which remain highly uncertain for most VOCs. The current state of knowledge of atmospheric chemistry as it relates to VOC reactivity is summarized below, and discussed in more detail in the Atmospheric Chemistry chapter in this assessment. The chemical mechanisms currently used or being developed for VOC reactivity are then summarized, and our ability to model ozone reactivities of various VOCs are discussed. Although both gas-phase and heterogeneous processes are discussed, the focus of this discussion is mainly mechanisms for prediction of ozone and other gas-phase products, since discussion of mechanisms for predictions of aerosol formation potential is given in a separate section, as discussed below. 3

73 Figure EX-1. Diagram of processes affecting the rate of ozone formation. The basic gas phase chemistry important in photochemical smog has been the subject of much study over the last fifty years. The oxidation of hydrocarbons begins with the abstraction of a proton by the hydroxyl radical (OH). In the presence of NOx (NOx = NO + NO2), the subsequent reactions result in the conversion of molecular oxygen to ozone. Ozone production continues as long as sufficient NOx is present so that reactions of peroxy radicals (RO2) with NO compete effectively with their reactions with other peroxy radicals. Ozone formation stops once NOx is consumed to sufficiently low levels. Conversely, high levels of NOx also inhibit O3 because reaction of OH with NO2 reduces OH levels. A diagram of the processes involved in ozone formation is shown in Figure EX-1. Another factor affecting the behavior of VOCs and NOx in ozone formation is competition for the hydroxyl radical. Although higher VOC concentrations generally lead to more ozone formation, increasing NOx for a given VOC concentration may increase or decrease ozone, depending on the prevailing VOC-to-NOx ratio. As a result, the rate of ozone production is not simply proportional to the amount of NOx present. The reader is referred to the Atmospheric Chemistry chapter for further details. The reactions that participate in smog formation can be divided into two groups: those that involve inorganic (non-carbon-containing) molecules such as OH and NO2, and reactions that involve VOCs and other organic species. Inorganic atmospheric reactions have been studied for many years, and the available laboratory data is evaluated on a regular basis. The reference documents produced by the NASA Stratospheric Data Panel (NASA, 1997) and the IUPAC Panel (Atkinson et al., 1997) are widely used as sources for the rate constants and other values 4

74 needed for modeling ozone formation. In general, gas-phase reactions of inorganic species are well characterized, but some are still under study to reduce the uncertainty associated with key interactions. The dominant area of remaining uncertainty with respect to inorganic species is that of reactions between gases and liquid or solid species. These heterogeneous reactions affect laboratory and atmospheric systems differently, and thus their characterization under both conditions is crucial to the extrapolation of smog chamber studies to the real atmosphere. In contrast to the subset of inorganic reactions, the organic reactions that contribute to local air quality are complex and still highly uncertain. The least uncertain aspect of these reactions is the initial rates of VOC reactions with OH or NO3 radicals or with ozone. Modern rate constant measurements are often precise, and where individual rate constants have been measured they are often known fairly well. Nevertheless, the stated uncertainties in rate constants in the compilations are almost always 25% or greater. For most compounds it is usually not particularly difficult or costly to obtain these rate constants if no data are available. While methods exist for estimating rate constants for the reactions of VOCs with OH and NO3 radicals when data are not available, these estimates are probably only good to a factor of two for most VOCs. Much of the complexity in organic photooxidation mechanisms comes from the variety of reactions in which the intermediate alkoxy radicals can participate. Absolute rate constants have been measured for only a few of the simplest alkoxy radical reactions, and most of the other available data concerns ratios of rate constants which can be inferred from results of product studies. This type of information is becoming available for an increasing number of systems because of ongoing product studies, though these branching ratios still need to be estimated for the large majority of VOCs emitted into the atmosphere. In recent years the development of theoretical methods for the calculation of potential energy surfaces allows the direct computation of some rate constants. These have not yet been exploited to any significant extent in the uncertain areas of atmospheric chemistry, and the time seems right for a serious theoretical look at many of these processes. Estimates of heats of reaction are also used in many of the estimation methods, and often can be used to rule out chemically unreasonable reaction schemes. Theoretical calculations could potentially be very useful in providing the data needed to support application of thermochemically-based estimation methods. The combined knowledge of atmospheric chemical processes can be used to estimate ozone production from a given release of VOCs. To do so, a chemical mechanism is developed and incorporated into an airshed model. Because of the large number of compounds emitted or formed in the polluted troposphere and the large numbers of reactions they, and their reactive products, can undergo, these mechanisms must necessarily contain significant simplifications and approximations. Furthermore, because of limitations in our knowledge, these mechanisms must contain assumptions and extrapolations to represent processes that are important but whose details are unknown. Different mechanism developers can apply different approaches to simplify or condense the mechanism to make it tractable and can use different assumptions and extrapolation methods when representing the main areas of uncertainty. 5

75 Different methods for mechanism compression include those based on reactivity, concentration weighting, and reactivity across carbon bonds within the molecules of each individual species. One alternate approach is to compress the mechanism in a transient fashion; the detailed speciation is retained until chemical integration is required, then the mechanism is compressed for the purposes of integration. After integration, the original information may be recovered. The development of these methods will allow increased hydrocarbon speciation in future modeling of reactivity. At the same time, increased model speciation in the absence of laboratory-based mechanistic or kinetic data will add little confidence to model results. The extent to which the use of detailed, temporarily compressed mechanisms improves ozone simulations has yet to be determined and would be an area worthy of further study. Most, though not all, of the mechanisms used in the current generation of models have been summarized by Bergin et al. (1997) and Dodge (1999). The major mechanisms relevant to current reactivity assessments are discussed in the Atmospheric Chemistry chapter. The Carbon Bond IV mechanism is important because it is widely used in regulatory models. Its rate constants and reaction schemes represent the state of knowledge as of approximately 1987, although some important rate constants have been updated since then. The RADM-2 mechanism developed by Stockwell et al. (1990) is used in the EPA s Regional Acid Deposition (RADM) model and is the only mechanism currently incorporated in the EPA s Models-3 system. This mechanism was recently updated, expanded, and renamed RACM ( Regional Atmospheric Chemistry Mechanism ). It is the most updated of the published mechanisms in terms of its rate constants and the mechanisms for its explicit reactions. The SAPRC mechanisms are important because they are designed specifically for VOC reactivity assessment and have been employed to generate reactivity scales that have been or are being considered for use in regulatory applications. The SAPRC-90 mechanism has been evaluated against results of ~500 smog chamber experiments and in most cases fits the ozone data to within ±30%. Alternatives to the SAPRC mechanism (e.g., the Master Chemical Mechanism (MCM) of Derwent and co-workers) are being developed and applied by researchers in Europe. These mechanisms are not used in the United States because model software is not adapted to mechanisms of this size. Also, they have not yet (to our knowledge) been evaluated against results of environmental chamber experiments. In Canada, the ADOM-II mechanism is currently used for reactivity simulations, and a new mechanism is under development for gas-phase and particulate modeling in the AURAMS model. Tests to date have shown that the new mechanism provides a significant improvement in the ability to predict chamber data compared to the ADOM-II mechanism. Before any chemical mechanism is incorporated into an airshed model, it must be demonstrated to predict at least the major features of the VOC-NOx-air photooxidation process. The only practical means for doing this is to conduct experiments using an environmental chamber, also called a smog chamber, where the chemical processes of interest occur under controlled and well-characterized conditions. Although there are really no practical alternatives at the present time, use of environmental chambers for mechanism evaluation is not without significant problems. Some shortcomings of chamber experiments include the high VOC concentrations at which most experiments must be conducted, the occurrence of reactions on 6

76 chamber walls, the inability to track most intermediate products, the difficulty of characterizing illumination inside the chamber, and the lack of multi-day experiments representative of conditions downwind of pollution events. In summary, recent years have seen significant progress in our understanding and ability to model the gas-phase reactions of pollutants in the troposphere. Nevertheless, there remain major gaps in our knowledge of the details of these gas-phase reactions, and our understanding of potentially significant heterogeneous processes is even less complete. Many of the mechanisms used to predict ozone are highly parameterized and simplified, with empirical adjustments to fit chamber data, and with no reliable ability to predict impacts other than on ozone and perhaps overall OH radical levels. Ongoing improvements in airshed model hardware and software permit use of more detailed atmospheric mechanisms, which have the potential to give more accurate and comprehensive predictions for any VOC of interest. However, without knowledge of the mechanistic details, any predictions made using these detailed mechanisms may be no more reliable than those of the simplified and parameterized mechanisms currently in use. Therefore, the main factors limiting the chemical accuracy of current and future airshed models are limitations in our knowledge of atmospheric chemistry and limitations in the environmental chamber database needed to verify the accuracy of model predictions. Aerosol Forming Potential Urban fine particulate matter is comprised of a complex mixture of both primary and secondary organic and inorganic compounds and emanates from a wide variety of sources. An important component that can significantly contribute to the fine particulate burden, especially during severe urban smog episodes, is secondary organic aerosol (SOA). Like ozone, secondary organic aerosol results from the atmospheric oxidation of VOCs, but it is generally formed only from the oxidation of VOCs comprised of six or more carbon atoms. This is because oxidation products must have vapor pressures that are sufficiently low to enable them to partition into the aerosol phase. The atmospheric chemical reaction pathways of VOC molecules sufficiently large to lead to SOA are complex, and resulting oxidation products are both numerous and difficult to quantify analytically. As a result, it is currently not possible to determine the aerosol formation potential of individual VOCs strictly on the basis of atmospheric chemical reaction mechanisms. This means that SOA yields must be determined experimentally. As discussed in the Aerosol Formation Potential chapter of this assessment, SOA yields have been determined experimentally for many individual VOCs. The data obtained indicate SOA yields vary widely with conditions, which means that, as with ozone reactivity, no single aerosol formation potential can be associated with any given VOC. However, Odum et al (1996) found that the California Institute of Technology (Caltech) environmental chamber data could be fit by a semi-empirical gas/aerosol partitioning model which assumes that the mix of semi-volatile oxidation products from a VOC can be represented by two empirical products, 7

77 characterized by parameters α 1, K 1, and α 2, K 2. Here, the α s refer to the yields of the products, and the K s refer to their gas-aerosol partitioning coefficient, which is a function of the molecular weight, vapor pressure, and the activity coefficient of the product in the aerosol phase mixture. Such parameters have been derived for a variety of aromatic hydrocarbons and biogenic terpenes. This gas/aerosol partitioning model can be used to predict SOA yields in a chamber experiment or in the atmosphere from N Y = M o i=1 8 α i K i 1+ K i M o where Y is the aerosol mass formed from the VOC (in terms of mass aerosol formed per mass VOC reacted), M o is the mass of aerosol present, and N is the number of semi-volatile products, which is assumed to be 2 in the semi-empirical model. It must be noted that this theory assumes that secondary products are unable to form a solution with existing inorganic seed aerosol. Furthermore, most of the experimentally determined SOA yields have been measured at relative humidity (RH) less than 5%. At this level of RH the seed aerosol is dry and the resulting organic aerosol is water-free. Because organic products will likely be most soluble in their own liquids, SOA yields measured at essentially 0% RH can be expected to represent an upper limit to the aerosol partitioning that will result. Experiments are presently underway at the Caltech to measure SOA yields as a fraction of RH over realistic ambient RH ranges. Because of the difficulties in characterizing SOA on a molecular basis and considering the significant uncertainties associated with model input data, the treatment of SOA in current air quality models is limited. Seigneur et al. (1998) recently reviewed the treatment of SOA formation in current regulatory and research-grade air quality models, and the results are summarized in the Aerosol Formation Potential chapter of this assessment. Of the eight models that treat SOA formation, six use an approach assuming a single SOA yield from each VOC, which as indicated above is not consistent with chamber data. The GATOR model (Jacobson, 1997) describes gas-particle partitioning of about ten condensable and soluble organic species that, however, are not directly representative of ambient condensable reaction products. In the CIT model, SOA is modeled by mass transport between gas and particulate phases, which is governed by the same formulation as that for inorganic volatile compounds. Recent work by Strader et al. (1998) combines the organic-phase absorption approach of Odum et al. (1996, 1997a,b) with recent experimental data on saturation vapor pressures of organic compounds, and derived a model with six condensable products, with yield parameters determined on the basis of Caltech chamber data. This approach has been incorporated in the revised version of the UAM-AERO air quality model. This approach differs from that of Odum et al in that the partitioning coefficient parameters, K i, were derived from theory rather than fits to chamber data. However, at present both methods rely on empirical data to determine the partition parameters.

78 The Aerosol Forming Potential chapter also discusses data on ambient aerosol composition. Although the available data provide some information on the magnitude and structure of secondary organic aerosol, this information is incomplete and requires that significant assumptions be made in interpreting the ambient data. Our understanding of the chemical nature and sources of secondary organics would be significantly improved by the identification of suitable molecular tracers of organic aerosol formation pathways. Atmospheric chemical reaction mechanisms currently included in urban-and regionalscale models were designed for accurate prediction of ozone formation chemistry. Generally, these mechanisms do not account for the chemistry of organics in sufficient detail to predict the generation of semi-volatile products. Many of the higher molecular weight organics that are not important in ozone formation, but are sources of secondary organic aerosol, are not represented in current mechanisms. To predict SOA formation, chemical reaction mechanisms need to be expanded to include all the important SOA-forming organics and to include, to the extent known, a representation of the semi-volatile products formed from each parent compound. Evaluation of the organics that need to be included will be based on available chamber data, on estimated atmospheric oxidation mechanisms, and on the estimated vapor pressures of the oxidation products. While this expansion may not lead to substantially improved ozone prediction, it is essential to prediction of SOA formation. Emissions Data The emissions from the multitude of sources in the environment are critically important inputs in the airshed models that must be used to assess impacts or reactivities of VOCs. The amount of emissions determines the nature of the chemical environment in which the VOC reacts, and thus affects not only the absolute amount of ozone and other secondary pollutants formed, but also the incremental effects of any additional VOC on the formation of those pollutants (i.e., the VOC reactivities). Emissions data also impact directly on government policy. Changes in emissions levels over time reflect the effectiveness of government regulations, as well as changes in activity level. The accuracy and completeness of emissions databases are therefore of great importance. The creation of an emissions database takes place in several stages, with each stage having associated errors and uncertainties. From sources such as Moody et al (1995), the SPECIATE web site (Ryan, 1998), Moran et al (1997), and the AP-42 database (EPA, 1995), typical stages for determining anthropogenic emissions are as follows: Source profiles (which provide a breakdown of the total VOC emitted from a given source type into a detailed VOC fractionation) are constructed from measurements. A limited number of profiles are currently available (e.g. about 300 used in SPECIATE). Source types are categorized using Source Classification Codes (SCC), Area Mobile Source codes (AMS) and Standard Industrial Classification codes (SIC). Typical emissions processing systems have thousands of source classifications (Ryan, 1998). Each classified source type is assigned a source profile from the set of available profiles. 9

79 The total annual VOC emission for each source type is determined from the product of the appropriate emission factor, activity rate, and control factors (AP-42; EPA, 1995). Individual sources from the database are assigned total annual VOC emissions, source classification codes, and source profiles. The total annual VOC emission rates may be multiplied by meteorological factors (e.g. to include dependence of emissions on ambient temperatures). The source profiles, linked to the individual sources in (5), are used to convert the total annual VOC emission into a speciated annual VOC emission. The detailed speciation is lumped into a smaller number of species for modeling purposes (e.g. Middleton et al., 1990). Temporal allocation occurs; temporal factors are applied to the model species annual emissions to create seasonal, daily, then hourly emissions estimates. Spatial allocation factors are used to create the spatial distribution of area sources. Smaller point sources are sometimes included with the area sources. For biogenic emissions, the processing differs somewhat: Spatially varying biomass factors (mass of emitting foliage / unit horizontal area) are linked with monthly frost/seasonal variation factors and a database of vegetation categories. Net vegetation standard emission rates (30 o C, 1000 µmol/m 2 /sec photo-synthetically active radiation (PAR)) are calculated for each part of the vegetation grid. Temperature and / or PAR corrections are applied for the meteorology at the site. For both forms of emissions, regridding from the database spatial grid to the model grid is often required. Each of the stages listed above has uncertainties, many of which have not been characterized in a quantitative way. The limited number of source profiles compared to SCC, SIC, or AMS codes is an example. The extent to which the available source profiles adequately represent the much larger set of source classifications is unknown. The source profiles themselves are the result of a relatively small number of measurement studies; in some cases a single study is used to represent the emissions from an entire industry. Errors in the total annual VOC emission rates are usually not estimated. Similarly, errors in temporal emission factor estimates are usually not estimated. For the biogenic emissions, the existence of detailed vegetation databases is required; these are not always available with sufficiently fine spatial detail for adequate resolution of emitting regions. Some of the sources of uncertainty have been characterized. The effect of lumping detailed speciation emissions into model speciation has been examined by Middleton et al (1990) - comparisons there and in later work (Kuhn et al., 1998; see also Lumping in the Atmospheric Chemistry chapter of this assessment) indicate that the effects of speciation in model ozone 10

80 production may be a smaller source of error than other model approximations. Similarly, the procedures for determining standard emission rates for specific vegetation types are well established and the uncertainties associated with these measurements are known (c.f. Guenther et al., 1996). From the standpoint of VOC reactivity and ozone formation, it is essential to determine which emissions-related uncertainties have the biggest impact on the predicted reactivity of the atmosphere. This in turn could be used to design measurement initiatives directed at reducing the most significant uncertainties. Sensitivity studies of model results associated with errors in the temporal factors, source profile assignment and total VOC emission rates should be directed towards determining which aspects of the emissions system have the biggest impact on reactivity. A few preliminary studies of this nature are discussed in the Emissions Uncertainty section of the Emissions Data chapter of the assessment. Emissions uncertainties can contribute 30 to 50% of the overall uncertainty in O3 predictions, while the contribution to uncertainty in VOC reactivity may be smaller (8%). Certain aspects of anthropogenic emission factor calculation are described in the Status of Anthropogenic Emissions section of the emissions data chapter. There it is noted that all available data for point sources should be taken into account, for both minor and major emitters. In addition, time factors and other parameters have typically been estimated based on episode days; non-episode, non-attainment days may therefore be incorrectly represented in emissions databases. Biogenic hydrocarbon emissions are described in a separate section of the chapter on emissions data. Note that the need for improved vegetation databases is given as the most significant need for further research. Emission reduction effects such as volatility suggest that some AP-42 emissions estimates may not adequately describe the amounts of VOCs that ultimately enter the atmosphere. These arguments are discussed below and in the Emissions and Volatility chapter of this assessment, since they are relevant to assessing reactivities of individual VOCs as well as the quality of the overall emissions inventory. Volatility and Fate The formation of tropospheric ozone from a volatile compound is a dynamic multi-step kinetic process that is highly dependent upon the relative concentrations of NOx and the VOC. The tropospheric concentration of a VOC is likewise affected by both the rate and extent of release from an emission source and by the rate of removal through a variety of competing processes (photooxidation, deposition, horizontal and vertical transport, aerosol formation) (Mackay et al, 1996). The latter considerations are clearly important in assessing effects of VOCs on formation of ozone and other impacts which involve the VOC reacting in the gas phase. These are discussed in more detail in the Volatility and Fate chapter of this assessment, and are summarized below. Consideration of equilibrium vapor pressures and partition between gas phase and aerosols suggests that many low-vapor-pressure (LVP) VOC will be present predominately in the gas phase. However, this conclusion may not be entirely correct in view of those studies 11

81 indicating alternate fates for the VOCs contained in consumer, commercial, and agricultural products (Rapaport, 1988; Bennett et al, 1998). There are many ways that compounds of low volatility, especially those that are hydrophilic, may be prevented from entering the atmosphere or removed once they enter, but quantitative assessments are rare. The rate of volatilization may dictate the overall rate of ozone formation for a particular VOC. Under some circumstances, the rate of release of a VOC to the environment may be the rate-limiting step in ozone production, thereby restricting its relative importance and overall contribution. Although, research has shown that the rate of volatilization from wastewater s and agricultural lands can affect airborne exposures to VOCs, there has been no systematic study of the volatilization rates from complex mixtures such as coatings, composites, and consumer products (Bianchi and Varney, 1977; Sharma and Overcamp, 1996). The overall impact of VOCs from these emission categories can be affected by the nature of matrix in which there are dissolved with lower rates of volatilization occurring when water soluble compounds are dissolved in an aqueous matrices (Lyons, 1990). Another important aspect of the volatilization issue is the role of competing removal processes. When volatilization rates are low, there is time for competing processes to remove the VOC by alternate routes (Watson et al, 1998). New research is needed to examine the issue of volatilization rate and to the develop models that can incorporate all that that is known about a chemical s tropospheric fate and removal. Because LVPs fall outside normal analytical methods, little is known about the ambient air concentration of semi-volatile compounds. Difficulties with source apportionment by chemical mass balance and analytical determination hinder any reliable estimation of the emissions rate for low volatility compounds. The total emissions of these compounds is not known with certainty, and the contribution from coatings and commercial products may be much lower than the proportions attributed to these sources in current emissions inventories (Watson et al, 1998). Air Quality Models As indicated above, because a VOC s impacts on air quality depend on the environment where it is emitted, use of airshed models provides the only practical means to predict the VOC s impact in an actual environment. Air quality models (AQMs) are computerized representations of the atmospheric processes responsible for air pollution, including ozone formation. The models simulate the atmosphere in varying degree of detail by mathematically representing emissions; initial and boundary concentrations; chemical reactions of emitted species and their products; and local meteorology such as sunlight, wind, and temperature. In this way, an understanding of the atmosphere's chemistry and meteorology is combined with estimates of source emissions to predict the possible effects of control strategies. AQMs are also an important tool in gaining understanding about the behavior of various compounds in the atmosphere, such as the reactivity of VOCs, and AQM simulations can be used to develop attainment strategies that minimize the costs of control strategy implementation. These models are essential for evaluating control strategies aimed at meeting air quality goals. 12

82 While AQMs are the best tools currently available for evaluating proposed ozone control strategies, it is important to recognize that the uncertainties in model components and in input data used by the models might have a serious impact on the model predictions. Significant progress has been made on uncertainty analysis, particularly with regard to uncertainty reduction in reactivity-based analysis methods. However, some serious concerns remain regarding the implementation of reactivity-based policy. Important issues in the application of AQMs for control strategy evaluation and reactivity quantification are as follows: Although serious questions still exist about the validity of models for the simulation of particular scenarios, it can be argued that we may have greater confidence in their ability to predict relative changes, such as for reactivity quantification. Box models and trajectory models are useful for calculating preliminary reactivity scales and for detailed uncertainty and sensitivity analysis. Box/trajectory models, however, are overly simplistic for evaluating reactivity-based substitution strategies and regional or national reactivity scales. Grid models should be used for these purposes. (This is discussed further in the Environmental Conditions section.) The traditional model evaluation criteria currently in use by the U.S. EPA are inadequate for determining the fitness of a model simulation for assessing control strategies. Various diagnostic approaches exist that may be useful in improving model evaluation, and these approaches should be considered in developing new criteria for model evaluation. For applications of models to reactivity scales, it is particularly important to develop diagnostic tests that validate model representation of odd nitrogen and radical budgets. As discussed in the Environmental Conditions chapter, the representation of transport and dispersion affects precursor concentrations and O 3 concentrations. This can also affect VOC to NO x ratios and O 3 sensitivity to precursors. While box models have overly simplistic dispersion schemes, the more physically complete grid models still have large uncertainties in their representation of vertical dispersion and horizontal advection. Uncertainty in transport processes such as nocturnal jets could significantly affect model predictions of reactivity. Sensitivity studies are needed to examine the possible effects of uncertainty in the meteorology on simulations of reactivity-based substitution strategies. Uncertainty and sensitivity analysis of model performance and reactivity prediction is crucial. Four types of sensitivity analyses are typically applied: 1. Stand alone analysis of the effect of inputs or model formulation. 2. Evaluation of outcome variable responses caused by a change in the process parameterizations, model structure/numerics, or input variables (e.g., the effect of a change in meteorology on O 3 concentration predictions.) 3. Evaluation of the change in outcome variable sensitivity to precursor emissions caused by a change in inputs (e.g., the effect of a change in meteorology on O 3 sensitivity to VOC reductions.) 4. Evaluation of the change in an emissions control strategy (magnitude or type) needed to attain an air quality standard. 13

83 It is necessary to determine the adequacy of each of these types of analyses in assessing reactivity-based substitution strategies. Additional analysis tools should be explored and current techniques should be applied to a wider range of parameters. Environmental Conditions Even if there were no uncertainty in the chemical mechanism or the airshed model when applied to a particular episode, the reactivity calculated for the VOC will be applicable only for that episode, and may not accurately represent impact in another area, or under different meteorological conditions. Nevertheless, most practical reactivity-based regulatory strategies being proposed involve using a single reactivity scale that will be applied on a national or at least regional basis. At best, the distribution of VOC reactivities in an appropriate ensemble of environmental conditions can be used as the basis for the regulatory decision. In either case, an assessment of the most appropriate type of reactivity scale, or set of scales, for regulatory applications requires knowledge of the distribution of environmental conditions relevant to the application of the regulation. In the Environmental Conditions chapter of this assessment, the environmental factors affecting reactivity, atmospheric structure, and potential impacts of long range transport, and variability in current reactivity scales are discussed. As discussed there and in the Reactivity Assessments chapter (see below), current urban reactivity scales presume that air near the surface moves slowly and is confined to the boundary layer and do not adequately consider long range transport mechanisms involving the coupling of the boundary layer to the free troposphere. For example, in many instances ozone and its precursors are rapidly vented aloft and quickly transported downwind of the mid-latitude jet stream, where they can be re-entrained into the boundary layer. It can be argued that reactivity scales that do not adequately account for long range transport mechanisms may not be appropriate for assessments relevant to the new ozone standard of 80 ppb averaged over 8 hours. Long-range transport of ozone and ozone precursors via the free troposphere has the potential to influence the level of radicals in remote urban and rural areas, thereby influencing the local reactivities of VOCs in those areas. Transport of ozone precursors has the potential to influence the NOx budget in remote rural areas, through reservoir species such as PAN and other organic nitrates that do not easily rain out of the atmosphere. It also has the potential to produce significant quantities of ozone in remote urban and rural areas when compared to the difference between the new federal ozone standard (80 ppb) and the level of background tropospheric ozone (30 ppb). Regional, 3-D Eulerian models that adequately resolve both the boundary layer and the free troposphere are presently the only tools that are believed to properly account for long-range transport mechanisms involving boundary layer-free troposphere coupling. Therefore, it is recommended that use of such models in reactivity assessments be expanded. This is discussed further in the following section. 14

84 Reactivity Assessments Over the past decade, studies have been performed to assess the reactivity of organics and the viability of using VOC relative reactivity rankings in emissions regulations. Both experimental and modeling based approaches have shown a wide range of reactivities, and, to a lesser extent, increased the availability of tools to quantify reactivity. While the early studies utilized smog chambers and box/trajectory model calculations, more recent studies have included three-dimensional photochemical modeling and applied uncertainty analyses to both box model and three dimensional assessments. The first studies conducted by Carter (e.g., Carter, 1994) showed that there is a wide variation in the ozone forming potential of VOCs and that the absolute amount of ozone formed per mass of VOC emissions differed significantly depending upon ambient conditions. For example, the absolute amount of ozone formed for virtually all species was found to decrease when NOx availability was reduced (e.g., MOIR vs. MIR conditions). However, the relative amount of ozone formed (e.g., the amount of ozone formed from 1 kg of emissions of the target species compared with the amount of ozone formed from 1 kg of a base mixture) was found to be a more stable and robust measure of VOC reactivity than an absolute measure, regardless of ambient conditions. As shown in Figure ES-2, the relative reactivity is also reasonably consistent regardless of modeling methods or quantification metric. Indeed, studies using both box and airshed models indicate that relative reactivity is significantly less sensitive to model uncertainties and environmental variabilities than are absolute measures. Uncertainties and variabilities led to relative reactivity coefficients of variation of 10-30% in most cases, suggesting that relative reactivity is viable beyond single airsheds or states. This hypothesis should be further investigated. At present, reactivity studies can be performed at both regional and urban scales. Threedimensional models, using advanced chemical mechanisms and sensitivity analysis techniques, have been applied to the areas of Los Angeles and the Texas-Mexico border to quantify and assess the spatial variability of relative reactivities of various VOCs. Similar modeling approaches have been applied at larger scales, e.g., the eastern United States and part of southeastern Canada, though not to quantify VOC reactivity. Such a spatial extension would be straightforward and could be used to further understand regional reactivity issues. As discussed in the Atmospheric Chemistry chapter, one issue that contributes significantly to the uncertainty in reactivity quantification is that chemical oxidation pathways are well known for only a relatively small number of compounds. Most of the VOC reaction paths are estimated. Reactivity estimates for novel compounds would be subject to high levels of uncertainty until their chemistry was elucidated. However, current Monte Carlo and sensitivity analysis techniques can establish bounds on reactivity estimates and aid in identifying the most important areas for further investigation. Studies to date suggest that, for the most part, the major uncertainty in relative reactivities is from uncertainty in the initial reaction rate constants (e.g., the reaction of the VOC with the hydroxyl radical). 15

85 Figure ES-2. Comparison of three-dimensional and trajectory modeled relative reactivities. Relative reactivity is defined as the ratio of the incremental reactivity of the VOC divided by the incremental reactivity of the mixture used to represent reactive VOC emissions from all sources CIT Peak Ozone CIT Population Threshold Exposure CIT Spatial Threshold Exposure Box Model MIR Box Model MOIR Relative Reactivity carbon monoxide ethane benzene methyl-t-butylether 2,2,4 trimethylpentane butane methanol methyl ethyl ketone 2-methylpentane ethylbenzene toluene ethanol ethyl-t-butylether methylcyclopentane 2-methyl-1-butene o-xylene 2-methyl-2-butene ethene 3-methylcyclopentene m,p-xylene acetaldehyde 1,2,4 trimethylbenzene propene isoprene propionaldehyde+higher 1,3 butadiene formaldehyde A strong foundation for understanding VOC reactivity issues has been developed, and the appropriate tools exist with which to further this understanding. Specific research projects should be identified to: enable the development of reactivity scales for use at the national levels (e.g., in the U.S., Canada, or Mexico) further define appropriate metrics for reactivity assessment apply prediction and analysis tools at larger scales and in a greater variety of locations and conditions, and further investigate the sources and impacts of uncertainties. 16

86 Persistent Organic Pollutants Persistent Organic Pollutants (POPs) are under increasing international scrutiny due to their potential for long-range transport and because of their possible impacts on the global environment. It has been proposed that organic compounds with even modest atmospheric lifetimes may accumulate in remote regions after being emitted into the air, through successive deposition to and re-volatilization from soil and water. On the other hand, slowly reacting compounds tend to have the least adverse impact near the environment where they are emitted, and reactivity-based control strategies may tend to encourage the use of such compounds. Since slowly reacting compounds tend to be the most persistent, this is an obvious concern when implementing such policies. A discussion of persistent organic pollutants and related issues is beyond the scope of the present assessment. It is recommended that an assessment of the current state of knowledge in this area be included as part of a comprehensive overall assessment of VOC reactivity. Appropriate independent expert(s) will need to be identified to carry out this assessment. 17

87 SUMMARY OF RECOMMENDATIONS The recommendations for research in the various areas covered by this assessment are summarized in this section. They are given for the various subject areas discussed above, though a number of recommendations span several of these areas. Note that there has been no attempt at this point to prioritize these research needs. This will be done when the overall RRWG research plan is formulated, which will take into account the results of the assessment of the relevant policy issues as well as the results of this science assessment. Atmospheric Chemistry Recommendations for research in the area of Atmospheric Chemistry and Chemical Mechanisms fall into three broad categories: laboratory studies to provide a more complete understanding of the processes which must be included in models of tropospheric reactivity, theoretical calculations and model improvements to allow for maximum use of our knowledge and performance of reactivity assessments, and environmental chamber improvements and validation experiments to provide a high-quality data base against which to test airshed and reactivity models. These are summarized below. Laboratory Studies Rate constants and mechanisms for reactions of peroxy radicals with NO, HO 2, other RO 2, and NO 3 radicals need to be studied. This would include additional data for nitrate yields from peroxy + NO reactions, particularly for non-hydrocarbon reactions. Branching ratios for the competing reactions of alkoxy radicals, particularly those not formed from alkanes and alkenes, need to be studied. Details of the reactions of ozone with alkenes and other VOCs containing double bonds under atmospheric conditions need to be elucidated. Total radical yields are particularly important in model simulations of VOC reactivity. More information is needed concerning the tropospheric chemistry of the oxygenated products formed in the photooxidation of VOCs. Laboratory and ambient air measurements need to be carried out to determine the atmospheric fate of potentially reactive oxygenated species, particularly those with a high OH reactivity (or expected high OH reactivity) and a known or expected low volatility. Further study of heterogeneous reactions in the atmosphere and in environmental chambers are needed. This includes, but is not limited to, reactions involving HONO formation, N 2 O 5 hydrolysis, and reactions involved in secondary aerosol formation. Studies are needed to improve our understanding of reactions on smog chamber walls. Such information will reduce uncertainties and sources of errors when using environmental chamber data to evaluate and develop chemical mechanisms. 18

88 More information is needed concerning heterogeneous processes in aqueous media. Needed studies include improved characterization of aerosol phase and composition, heterogeneous chemistry of peroxy radicals, HONO as an early morning HO x source in urban areas, temperature dependence of aqueous phase reactions, and compilation of data for modeling atmospheric aqueous phase chemistry. Information is needed concerning possible reduction of HNO 3 to NO x on soot or sulfate aerosol. Reactivity of O 3 on soot, mineral dust, and organic carbon aerosols is also poorly known at this time. Mechanisms and products of the reactions of OH-aromatic adducts with O 2 and NO 3 remain highly uncertain. Quantitative yield information and studies of the reactions (including photolyses) of these aromatic products are especially needed. Thermal decompositions and other atmospherically important reactions of the higher PAN analogues, such as that formed from methacrolein and isoprene, need to be studied. Information concerning the reactions of radicals formed from halogen-containing compounds, amines, and other nitrogen-containing compounds is needed before reactivities of such compounds can be assessed with any accuracy. Computational and Modeling Techniques The computation of rate constants using ab initio, transition state, and other theoretical methods needs to be applied to atmospherically-relevant systems. These techniques should be tested through comparison to known processes, then applied towards predicting mechanisms and reaction rates which are currently unknown. The methods used to estimate mechanisms of VOCs for which there are no data need to be improved. For many classes of compounds no reliable estimation methods yet exist. Theoretical calculations of the most uncertain reactions and targeted experimental studies to establish or evaluate relationships are needed. Sensitivity analysis should be applied to the mechanisms to help decide which processes will be most important to study in the future. This requires quantifying the uncertainties involved, not only in the elementary rate constants, but also in parameterization methods in mechanisms adjusted to fit chamber or other data. Work is needed to assess the optimum level of detail for atmospheric chemical mechanisms, given the modeling application and knowledge of the processes being represented. The minimum list of inorganic reactions required for photochemical reactivity calculations, including the pressure and water vapor dependent pathways of HO 2, needs to be determined. A model comparison of highly speciated versus lumped versus temporarily compressed mechanisms should be performed for a realistic atmospheric conditions to determine the relative merits of model speciation in reactivity estimates. 19

89 The implementation of the Morphecule approach needs to be completed, and its advantages over alternative methods for representing chemical detail in models need to be assessed. Smog Chamber Studies Improved facilities are needed to evaluate mechanisms under lower pollutant conditions than is currently possible, and improved instrumentation is needed to monitor trace species and intermediates and VOC reaction products. Methodologies need to be developed to screen or assess reactivity more readily and at lower cost than is currently is possible. Aerosol Formation Potential Research recommendations concerning assessment of aerosol forming potentials of VOCs can be summarized as follows: Aerosol formation yields need to be measured for the full variety of VOCs of interest. Such measurements need to be made under a sufficient variety of conditions to determine all relevant parameters that will affect SOA yields in the atmosphere, including temperature and humidity. The environmental chamber systems used to determine SOA yields need to be improved so that conditions affecting aerosol formation, particularly temperature and humidity, can be controlled and systematically varied. Experimental data are needed under controlled conditions to test theories and models for of SOA formation. The treatment of SOA formation in models needs to be improved. At a minimum, the treatment must be consistent with available environmental chamber data. Gas-Phase atmospheric chemical reaction mechanisms need to be adapted to predict formation of semi-volatile products from the major classes of emitted VOCs and to appropriately represent their reactions. The types of semi-volatile products formed from the reactions of the major classes of emitted VOCs need to be determined, and their vapor pressures and other parameters affecting their partitioning into the vapor phase need to be determined. More information is needed concerning the chemical composition of the organic components in ambient aerosol, and of organic aerosol formed from VOCs in environmental chamber experiments. 20

90 Emissions Data The following recommendations are made concerning the problem of reducing the uncertainty of the emissions data which are used as inputs to the air quality models used for reactivity assessment: The impact of the uncertainties and approximations used in the emissions assignment process on the results of models used for reactivity calculations needs to be determined. Sensitivity analyses should be conducted to identify the effect of uncertainties and approximations in temporal factors, source profile assignment and total annual VOC emission rates on model reactivity estimates. The biogenic emissions data are particularly uncertain, and these uncertainties may significantly affect reactivity assessments in regional scenarios. There is a need for the improvement of the vegetation land-use database and area average factors used in biogenic emissions estimates to reduce their uncertainty. Other significant data gaps include lack of factors genetic variability within species and for wounding of vegetation, inadequate characterization of emission underestimates due to chemical losses, and the need for refinement of emission activity and escape efficiency parameters. EIIP procedures, operational information, and stack sampling should be used to ensure adequate speciation of anthropogenic VOCs, with the encouragement of industry participation. Time factors for apportionment of VOCs on non-episode days need to be designed, as well as appropriate changes to engineering and VOC composition parameters to reflect non-episode emissions days. Updates to the limited set of speciation profiles are essential for improvement to the emissions databases. Volatility and Fate The following research recommendations are made concerning the issue of the extent to which individual VOCs are emitted into the gas phase and undergo the gas-phase reactions involved in the formation of ozone, SOA, and other secondary pollutants. Alternative Fate factors for Partitioning need to be designed, to account for alternative fates such as losses to waste-water, chemical transformations, adsorption onto sinks Volatility factors need to be designed to account for the effects of volatility on net emissions. More validation data are needed for fate and transport models. Methods need to be developed to incorporate the possibility of alternative fates besides gas-phase reaction in the models used to assess VOC reactivity. These are not represented in the EKMA models used to develop the MIR and other scales, and the Eulerian models generally only represent deposition of compounds in the mixed layer. 21

91 Fate and transport models need to be enhanced to incorporate gas-phase atmospheric reactions. The principal physicochemical properties used in environmental fate analysis need to be determined for a wider variety of VOCs. These include Henry s law constant, octanolwater partition coefficient, air-octanol partition coefficient, and vapor pressure at 25 C. These parameters are generally available for common volatile compounds, but the values for low-vapor-pressure compounds are often undetermined and must be estimated. Methodologies for the ambient air measurements need to be improved so a wider variety of VOCs, particularly semi-volatile oxygenated compounds, can be monitored in the gas phase. The list of commonly measured substances in ambient air does not include semivolatile oxygenated compounds which are in the emissions inventory. More research on VOC monitoring methods is needed to ensure data reliability. Field studies designed to measure the emitted species are required to test any modeled hypothesis regarding atmospheric fate of emitted species,. These are a necessary last step in determining the accuracy of both the emissions profiles and the assumptions made with regards to the relationship between usage, emissions and ambient air concentrations. Experimental costs may require the generalization of a limited number of measurements to a larger class of compounds or emissions types. Sensitivity Analysis studies are needed to examine how the ozone yield from VOCs can be affected by factors affecting the VOC volatilization and removal from the troposphere. These studies are needed using urban, regional, and global models that include state of the science chemical mechanisms for VOC photooxidation. Air Quality Models Recommendations for future research concerning the air quality models used to determine VOC impacts in ambient environments are as follows: Appropriate model evaluation criteria need to be developed for examining reactivity and other control strategies. The available tools for sensitivity, uncertainty, and reactivity analyses need to be evaluated, and more appropriate tools may need to be developed. Methods need to be developed to aid in incorporating available measurement data into models, to reduce uncertainty in reactivity predictions. Studies are needed to examine the possible effects of uncertainties on three-dimensional simulations of reactivity-based substitution strategies. Environmental Conditions The following research areas are recommended to address the issues raised in the assessment of the representation of environmental conditions. Note that some of these 22

92 recommendations involve improvements in airshed model formulation, reactivity assessment methods, or in the chemical mechanisms, and thus could have been included in those sections. The role of upper and lateral boundary conditions for ozone and ozone precursors in determining the reactivities of VOCs in urban and rural airsheds needs to be investigated. Models with improved representations of convective, mesoscale, and synoptic-scale transport in large-scale reactivity models, including deep convection, frontal motions, and tropopause folding events need to be used in reactivity assessments Chemical mechanisms used in reactivity assessment models need a more complete representation of radical source, including chlorine radicals from sea salt in coastal airsheds. Chemical mechanisms used in reactivity assessments need better resolution of organic nitrates, including both soluble multi-functional organic nitrates that may deplete NO x from polluted air masses and non-soluble organic nitrates that can deliver NO x to remote rural environments. Rigorous measurements of the total reactive nitrogen budget need to be made in urban and rural areas within the United States to provide more extensive ground truth for both chemical mechanisms and airshed models. More emphasis on evaluating export-oriented approaches to reactivity is needed. Large-scale 3-D modeling assessments of the effectiveness of existing and alternative reactivity-based control strategies are needed. This is discussed further below. Reactivity Assessments In addition to the above recommendations concerning improved models and chemical mechanisms, the following recommendations can be made concerning reactivity assessments in general. Assessment of Relative Reactivity need to be made on a regional scale. To date, studies have primarily concentrated on using three-dimensional models to assess VOC reactivity at the urban (Los Angeles) and small regional (e.g., Texas) scales. This type of study should be expanded to assess reactivities at large regional scales (e.g., eastern United States, Eastern Canada). Further, the spatial variability in reactivity should be quantified for important VOCs. Determination Appropriate Reactivity Scale Metrics need to be determined. To date, there has been little investigation on the variety of reactivity scales that can be defined. This may involve assessments of health effects, economic impacts, and other policyrelated issues. 23

93 Persistent Organic Pollutants No specific recommendations are made concerning research areas related to persistent organic pollutants except that assessment of the current state of knowledge in the area of persistent organic pollutants as it relates to VOC reactivity needs to be carried out. 24

94 REFERENCES AP-42 database on Air Chief 4.0 CD-ROM, EFIG/EMAD/OAQPS/EPA, USEPA, 1995 Atkinson, R., D. L. Baulch, R. A. Cox, R. F. Hampson, Jr., J. A. Kerr, M. J. Rossi, and J. Troe (1997): Evaluated Kinetic, Photochemical and Heterogeneous Data for Atmospheric Chemistry: Supplement V., IUPAC Subcommittee on Gas Kinetic Data Evaluation for Atmospheric Chemistry, J. Phys. Chem. Ref. Data, 26, Bennett, D. H., T.E. McKone, M. Matthies, and W.E. Kastenberg (1998). General formulation of characteristic travel distance for semivolatile organic chemicals in a multimedia environment. Environ. Sci. Technol. 32, Bergin, M.S., Russell, A.G., Carter, W.P.L., Croes, B., and Seinfeld, J.H. "VOC Reactivity and Urban Ozone Control." Encyclopedia of Environmental Analysis and Remediation, pp J. Wiley & Sons, New York, NY.; Bianchi, A. P. and M.S. Varney (1997). Volatilisation processes in wastewater treatment plants as a source of potential exposure to VOCs Ann. Occup.Hyg. 41, 437. Carter, W.P.L. (1994). J. Air and Waste Mgmt. Assoc., 44: Dodge M.C., (1999): Chemical oxidant mechanisms for air quality modeling: Critical review paper for 1998 ozone assessment (to be submitted to Atmospheric Environment, Available at Guenther, A., W. Baugh, K. Davis, G. Hampton, P. Harley, L. Klinger, L. Vierling, P. Zimmerman, E. Allwine, S. Dilts, B. Lamb, H. Westburg, D. Baldocchi, C. Geron and T. Pierce, Isoprene fluxes measured by enclosure, relaxed eddy accumulation, surface layer gradient, mixed layer gradient, and mixed layer mass balance techniques, J. Geophys. Res., 101, , Jacobson, M. Z. (1997): Development and application of a new air pollution modeling system. Part II: Aerosol module structure and design, Atmos. Environ., 31, Kuhn, M., P.J.H. Builtjes, D. Poppe, D. Simpson, W.R. Stockwell, Y. Andersson-Skold, A. Baart, M. Das, F. Fiedler, O. Hov, F. Kirchner, P.A. Makar, J.B. Milford, M.G.M. Roemer, R. Ruhnke, A. Strand, B. Vogel, and H. Vogel, Intercomparison of the gasphase chemistry in several chemistry and transport models, Atm. Env., 32, , Lyons, D. (1990). Computer modelling of solvent blends. J. Oil Colour Chemists Assoc. 73, 82. Mackay, D, A. Di Guardo, S. Paterson, G. Kicsi, and CE. Cowan. (1996). Assessing the fate of new and existing chemicals: A five-stage process. Environ. Toxicol. Chem. 15,

95 Middleton, P., W.R. Stockwell, and W.P.L. Carter, Aggregation and analysis of volatile compound emissions for regional modelling, Atm. Env., 24, , Moody, T., J.D. Winkler, T. Wilson, S. Kersteter, The development and improvement of temporal allocation factor files, EPA report EPA-600/R , USEPA Office of Research and Development, Washington, D.C , Moran, M.D., M.T. Scholtz, C.F. Slama, A. Dorkalam, A. Taylor, N.S. Ting, D.Davies, P.A. Makar, and S. Venkatesh, An overview of CEPS1.0: Version 1.0 of the Canadian Emissions Processing System for regional-scale air quality models. Proc. 7 th AWMA Emission Inventory Symp., Oct , Research Triangle Park, North Carolina, Air & Waste Management Association, Pittsburgh NASA (1997): Chemical Kinetics and Photochemical Data for Use in Stratospheric Modeling, Evaluation Number 12, JPL Publication 97-4, Jet Propulsion Laboratory, Pasadena, California, January. Odum, J. R., Hoffmann, T., Bowman, F., Collins, D., Flagan, R. C., and Seinfeld, J. H. (1996) Gas/particle partitioning and secondary organic aerosol yields. Environ. Sci. Technol., 30, Odum, J. R., Jungkamp, T. P. W., Griffin, R. J., Forstner, H. J. L., Flagan, R. C., and Seinfeld, J. H. (1997a), Aromatics, reformulated gasoline, and atmospheric organic aerosol formation, Environ. Sci. Technol., 31, (1997). Odum, J. R., Jungkamp, T. P. W., Griffin, R. J., Flagan, R. C., and Seinfeld, J. H. (1997b) The atmospheric aerosol-forming potential of whole gasoline vapor. Science, 276, Rapaport, R. A. (1988). Predicition of consumer product chemical concentrations as a function of publicly owned treatment works treatment type and riverine dilution. Environ. Toxicol. Chem. 7, 107. Ryan, R., EPA Website [ accessed Nov 23, Sharma, M. and T.J. Overcamp (1996). Air emissions of volatile liquids spilled on sands. Environ. Prog. 15, 229. Seigneur, C., Pai, P., Louis, J. F., Hopke, P., and Grosjean, D. (1998) Review of threedimensional air quality models for particulate matter, Report No. 4669, American Petroleum Institute, Washington, D.C. Stockwell, W. R., P. Middleton, J. S. Chang, and X. Tang (1990): The Second Generation Regional Acid Deposition Model Chemical Mechanism for Regional Air Quality Modeling, J. Geophys. Res. 95,

96 Strader, R., Gurciullo, C., Pandis, S., Kumar, N., and Lurmann, F. W. (1998) Development of gas-phase chemistry, secondary organic aerosol, and aqueous-phase chemistry modules for PM modeling, Draft Final Report STI DFR, Coordinating Research Council, Atlanta, Georgia. Watson, J. G., J.C. Chow, and E.M. Fujita (1998). Review of volatile organic compound source apportionment by chemical mass balance Ozone Assessment - Critical Review Paper, North Atlantic Research Strategy for Tropospheric Ozone, Pasco, WA. 27

97 LIST OF CONTRIBUTORS Given below is the list of people who contributed to the various sections of this draft of the assessment, along with their affiliations and the sections to which they contributed. The persons taking the lead in compiling the various sections are also indicated. Participants are listed in alphabetical order. D.Allen (University of Texas, Austin): Aerosol Potential M. Bergin (Originally at University of Colorado, now at Georgia Institute of Technology, Atlanta): Air Quality Models; Reactivity Assessments (co-leader) D. W. Byun (NOAA): Environmental Conditions W. P. L. Carter (University of California, Riverside): Overall team leader; Executive Summary; Atmospheric Chemistry; Reactivity Assessments D. R. Crosley (SRI International): Atmospheric Chemistry J. A. Dege (Dupont Co.): Environmental Conditions R. Derwent (Meteorological Office, Bracknell, UK): Air Quality Models D. M. Golden (SRI International): Atmospheric Chemistry R.W.. Hamilton (Amway Co.): Emissions and Volatility A. Hanna (MCNC): Air Quality Models L. T. Iraci (SRI International): Atmospheric Chemistry J. C. Johnston (SRI International): Atmospheric Chemistry J. J. Kurland (Union Carbide Co.): Emissions and Volatility P. A. Makar (Atmospheric Environment Service, Ontario, Canada): Atmospheric Chemistry; Air Quality Models; Emissions and Volatility (leader), J. Milford (University of Colorado): Reactivity Assessments M. D. Moran (Atmospheric Environment Service, Ontario, Canada): Emissions and Volatility D. A. Morgott (Eastman Kodak Co.): Emissions and Volatility 28

98 J. D. Neece (Texas Natural Resource Conservation Commission): Reaction diagram in Executive Summary E. P. Olaguer (Dow Chemical Co.): Air Quality Models; Environmental Conditions (leader); Persistent Organic Pollutants P. J. Ostrowski (Occidental Chemical Co): Reactivity Assessments; Environmental Conditions. P. Pai (AER, San Ramon, CA): Aerosol Potential A. Russell (Georgia Institute of Technology, Atlanta): Reactivity Assessments (co-leader); Emissions and Volatility, Air Quality Models J. H. Seinfeld (California Institute of Technology): Aerosol Potential (leader) G. Sistla (New York State Department of Environmental Conservation): Emissions and Volatility G. Tonnesen (U.S. EPA): Air Quality Models (leader) Z. Wang (University of North Carolina at Chapel Hill): Air Quality Models. 29

99 AIR QUALITY MODELS Draft as of 11/26/98 Gail Tonnesen 1, Jay Olaguer 2, Michelle Bergin 3, Ted Russell 3, Adel Hanna 4, Paul Makar 5, Dick Derwent 6, and Zion Wang 7 1 University of California at Riverside 2 Dow Chemical Co. 3 University of Colorado and Georgia Institute of Technology, Atlanta 4 MCNC 5 Environment Canada 6 Meteorological Office, Bracknell, UK 7 University of North Carolina at Chapel Hill CONTENTS INTRODUCTION: AIR QUALITY MODEL ASSESSMENT... 2 MODEL COMPONENTS... 3 Governing Equations... 3 Domain, Grid Structure and Resolution... 6 Initial and Boundary Conditions... 6 Emissions... 7 Emissions Processing Methods... 8 Meteorological Inputs... 9 Photolysis Rates...10 Gas Phase Chemical Mechanisms Mechanism Inter-Comparisons Clouds and Aqueous Chemistry Aerosol Dynamics Routines Vertical Transport and Diffusion Chemical Solvers The GEAR Solver Lower Accuracy Methods Mathematical and Computational Implementation UNCERTAINTIES IN MODEL INPUTS: IMPLICATIONS FOR MODEL RELIABILITY DIAGNOSTIC METHODS Brute Force Sensitivity Direct Sensitivity Analysis Process Analysis...27 Ozone Source Apportionment Technology Indicator Ratios

100 TYPES OF MODELS Trajectory and Box Models OZIP/EKMA The Harwell Trajectory Model AES Box Model EMEP Trajectory Model Urban/Regional/Multi-Scale Grid Models Global Models...34 REFERENCES TABLES FIGURES INTRODUCTION: AIR QUALITY MODEL ASSESSMENT Many variabilities are inherent in different urban airsheds, such as the natural background chemical composition, incident radiation from the sun, and the chemical profile and emission rates of local sources. As is well known, tropospheric ozone is not emitted, but is formed in the atmosphere through a complex, non-linear process involving oxides of nitrogen (NO x ), different types of volatile organic compounds (VOCs), and sunlight. For this reason, it is difficult to predict how the reduction of differing source emissions may effect the formation of ozone in a given location. Because of the financial commitment required to implement almost any pollution control strategy, the potential impact of any control strategy on the atmosphere must be as fully assessed as possible before one is selected. Air quality models (AQMs) are computerized representations of the atmospheric processes responsible for air pollution, including ozone formation. The models simulate the atmosphere in varying degree of detail by mathematically representing emissions; initial and boundary concentrations of chemical species; the chemical reactions of the emitted species and of their products; and the local meteorology such as sunlight, wind, and temperature. In this way, an understanding of the atmosphere s chemistry and meteorology is combined with estimates of source emissions to predict possible control strategy effects. AQMs are also an important tool in gaining understanding about the behavior of various compounds in the atmosphere, such as the reactivity of VOCs. Models can also be designed to optimize the costs of control strategy implementation. These models are essential to evaluating control strategies aimed at reducing pollution to meet air quality goals. While AQMs are the best tools currently available for evaluating proposed ozone control strategies, it is very important to recognize that uncertainties in the model components and in the input data used by the models can have a serious impact on the model predictions. Significant progress has been made on uncertainty analysis, particularly with regard to uncertainty reduction in reactivity-based analysis methods, however some serious concerns remain regarding the implementation of reactivity-based policy. 2

101 The following section gives an overview of the general components of AQMs, and discusses the uncertainties in model parameters and inputs and the diagnostic methods used to examine their effects. Then, current models used to examine reactivity are summarized, and model performance evaluation is discussed. Finally, the selection and definition of scenario conditions for reactivity studies are presented. MODEL COMPONENTS Governing Equations Meteorological models can be used to solve the equations of hydrodynamics to predict air density (ρ), wind velocity vector (V), energy (E), and pressure (P). The governing equations are provided by the laws of conservation of mass, energy, and momentum, and the ideal gas law. Conservation of mass provides the mass continuity equation: ρ + Vρ = 0 (1) t We can neglect molecular diffusion in Equation (1) because advective processes are much larger than molecular diffusive processes in the troposphere and stratosphere. Because trace species in the atmosphere (excluding water) have concentrations at or below the part per million (ppm) level, they do not affect the solution of the equations for the wind fields. This provides an important simplification for AQMs because it allows the mass continuity equations for trace species to be solved independently of the continuity equations for air. Thus, wind fields generated by meteorological models can simply be read as inputs for an AQM. For the purposes of AQMs, we can write the mass continuity equation for each trace species as: C + V C = 0 (2) t where C represents either the concentration (mass/volume) or number density (molecules/volume) of the trace species. Equation (2) is called the flux form, or conservation form, because it is written in terms of the mass flux (VC). We can rewrite Equation (2) in the advective form: C + V C = C V (3) t where the right hand side of Equation (3) would be zero in pressure based coordinate systems or if divergence and convergence of air masses were considered negligible. We can also define a dimensionless quantity called the mixing ratio of the trace species as: 3

102 u C (4) ρ Substituting Equation (4) into Equation (3) and rearranging provides: u u ρ + V u = + Vρ = 0 t ρ t (5) From Equation (1) we see that the right side of Equation (5) is zero, so the flux and advective forms of the continuity equation are equivalent when written in terms of mixing ratio. We can solve the continuity equation in terms of number density using either the flux form or in the advective form using mixing ratio, whichever is most convenient. Rood (1987) has discussed the advantages of solving the continuity equation in the flux form and advective form. It may be easier to avoid instability and to ensure conservation of mass with number density in the flux form. It is simpler to treat divergence and vertical transport in the advective form because the mixing ratio is not affected by changes in pressure or temperature. Byun (1998a&b) has examined the solution of the continuity equations and conservation of mass using generalized coordinate systems. The various forms of the advection equation discussed above represent the instantaneous continuity equation. Air motions, however, are generally turbulent. We can write the instantaneous number density and velocity as the sums of mean components and turbulent fluctuation components: and the mean flux can be written as: C = C + C (6) V = V + V (7) V C = V C + V C (8) Substituting the mean expressions into Equation (2) gives us the mean continuity equation: C t + ( C ) = ( V C ) V (9) where the right-hand side of Equation (9) represents dispersion caused by both small and large scale turbulent processes, or eddies. Turbulent dispersion introduces additional unknowns into the continuity equation. Closure theory is the large body of work which attempts to close the continuity equation by finding various theoretical or observational relations in turbulence to solve for the additional unknowns. One common approach is to treat eddy dispersion analogously to molecular diffusion by defining an eddy diffusivity coefficient (D). More 4

103 sophisticated approaches such as transillience theory () attempt to represent the transport between non-adjacent grid cells caused by large scale eddies. Using the eddy diffusivity approach, and dropping the bars for the mean values, we obtain the familiar form of the advection-dispersion equation: C t + V C = D C (10) Finally, trace species can be produced and consumed by chemical reactions, where the production rate (P) and loss frequency (L) of a given species may be non-linear and also vary as a function of the concentrations of other species within the system. Species concentrations may also be affected by other sink and source terms (S) such as emissions and deposition. Thus, an advection-dispersion-reaction (ADR) equation can be written for each species i: C t i + V Ci = D Ci + Pi( C) Li( C) Ci + Si for i = 1, N (11) where N is the number of species represented in the photochemical mechanism. Equation (11) produces a system of N non-linear, partial differential equations (PDE) that are coupled by the species concentration vector (C). Equation (11) cannot be solved analytically, but various numerical methods can be used to obtain approximate solutions. Numerical methods can be divided into two main categories: Lagrangian approaches and Eulerian approaches. Lagrangian models are often referred to as trajectory models because they simulate photochemistry in parcels of air that follow the wind trajectory. The frame of reference is defined relative to the wind vector, so the advection term drops out of Equation (11). If turbulent dispersion is also ignored, Equation (11) reduces to a system of ordinary differential equation (ODE): dc dt i = Pi( C ) L( C) Ci + Si for i = 1, N (12) and the problem reduces to the solution of a system of stiff ODEs. This greatly simplifies the numerical solution of the model because accurate, efficient GEAR solvers are available for systems of stiff ODEs (Gear, 1971). These are discussed below. Strengths and weaknesses of Lagrangian models are discussed below. Eulerian approaches divide the problem domain into a grid of discrete elements or cells. Operator splitting and various numerical methods are then used to solve Equation (11) over the full domain. Components of grid models are summarized below and specific grid models are described below. 5

104 Domain, Grid Structure and Resolution The first Eulerian AQMs were limited to the urban scale and used grid resolutions on the order of 4 km.. The realization that long-range transport of pollutants and their precursors can impact local control strategies and the need to study regional impacts (as well as rapidly increasing computational resources) led to the use of coarse-grid regional scale models (e.g., RADM, ROM) to define boundary conditions for urban scale AQMs. More recently, multi-scale models have been developed that nest several levels of increasingly refined grids within a single modeling system (e.g., MAQSIP, CMAQ). The simplest and most commonly used nesting approach is to define a set of uniform, rectangular grids over the model domain. This approach is still impractical to adequately resolve the fine scale structure in species distributions caused by intense point source emissions. Fine scale resolution of point source plumes has been achieved using nested plume-in-grid (PiG) modules. The simple nested-grid approach used in AQMs is relatively inefficient compared to other approaches widely used in computational fluid dynamics. Sophisticated systems exist for automatically creating and refining irregular grid systems using finite element methods. These approaches have not been widely used in AQMs, although one notable exception, the Urban-to- Regional Multiscale (URM) Model (Odman and Russel, 1991; Kumar and Russell, 1996), is discussed below.. It is possible to use relatively simple (albeit inefficient) rectangular grids in AQMs because the domain of interest is well defined and species concentrations are relatively well behaved (i.e., no sharp fronts typical of ground water problems or shock waves typical of physics problems). Grid resolution can affect model predictions by introducing artificial dilution caused by averaging point source emissions into coarse grids and by increased numerical dispersion in advection solvers on coarse grids. Increased artificial dilution would be expected to cause lower precursor and O3 peak levels but higher regional levels (Sillman et al., 1990). It is not clear what effect artificial dilution would have on calculations of reactivity. The response is difficult to predict because VOC/NOx ratios and predicted peak O3 concentrations would likely increase in some cells and decrease in others. Additional investigation is required to determine the effects of grid resolution on model-simulated reactivity applications. While urban scale domains may be adequate for rapidly reacting VOC, reactivity calculations for long lived species should be simulated on model domains sufficiently large to simulate the chemistry over a period of several days. Simulated incremental reactivity of NOx may be especially sensitive to grid resolution and the use of plume-in-grid modules. Initial and Boundary Conditions When a grid-based photochemical model is applied to simulate a pollution episode, it is necessary to specify concentration fields of all the species computed by the model at the beginning of the simulation. These concentration fields are called the initial conditions. 6

105 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: 1) use the output from a regional or global scale photochemical model; 2) use objective or interpolative techniques with ambient observational data; or 3) use default background values and expand the area that is modeled and lengthen the simulation period to minimize the uncertainties due to lack of measurements. The third technique is useful for areas sufficiently isolated from significant upwind sources. In the ideal case, observed data would provide information about the concentrations 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 or global 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 modeled region that concentrations approaching background values can be used for the upwind boundary conditions. Simulations of a multi-day pollution 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. The effect of uncertainties in boundary conditions on determinations of reactivity may be very important if the balance of NOx in the modeled region depends substantially on the influx of total reactive nitrogen from outside the region. Likewise, the total supply of radicals in the modeled region may depend critically on ozone advected downwards from the upper boundary, as discussed in the chapter on Environmental Conditions. Uncertainties in initial conditions, on the other hand, are not as likely to significantly impact estimates of reactivity compared to uncertainties in boundary conditions. Emissions A key use of air quality models is to determine how pollutant concentrations respond to emissions inputs, and accurate emissions inputs are key to good model performance. Emissions inputs are developed to be compatible with the chemical mechanism used in the model, and with the model resolution (both vertically and horizontally). Typically, this would include hourly, spatially gridded estimates of the emissions of CO, NO, NO 2, SO 2, SO 3 and the various primary VOCs in the mechanism. As discussed in the NARSTO summary, such emissions estimates are developed using emissions models, and like air quality models, they combine descriptions of various processes. 7

106 For example, such models may combine estimates of process rates (e.g., the amount of chemical produced in a plant) and emissions factors (e.g., tons of emissions per ton of processed chemical) to produce an estimate of the emissions from a single industrial facility. Another calculation may use the mileage traveled along a road in conjunction with the emissions per mile (accounting for the vehicle ages in the fleet, vehicle speeds and temperature), while a third would resolve the emissions associated with residential living (space heating) with the population distribution. Biogenic emissions are estimated from the type of vegetation and the emissions from that type of plant as a function of sunlight and temperature. Emissions inventories contain considerably more chemical detail than is typically used in condensed photochemical mechanisms so, as discussed below, the VOC emissions are lumped into the appropriate chemical mechanism categories. On the other hand, the available emissions information is often less detailed spatially than a model could use, being gathered on a county or state/province basis in many cases. In this case, the estimates must then be assigned to each grid using an appropriate surrogate (e.g., population weighting). The dominant emissions inventory preparation programs include the Flexible Regional Emissions Data System (FREDS) (Modica et al., 1985) and Emissions Modeling System-1995 (EMS-95) (Wilkinson et al., 1994) and the emissions preprocessor system (EPS). It is generally believed that emissions are one of, if not the most, uncertain inputs into air quality models. For example, results from the Southern California Air Quality Study suggest that the mobile source VOC exhaust emissions estimates may be low by a factor of two to four (Pierson et al., 1990). Biogenic emissions (of both VOCs and NO x ) are believed to be very uncertain as well, often suggesting an uncertainty of a factor of three or more (e.g., Geron et al., 1994; Simspon et al., 1995). Many other sources would also be susceptible to significant uncertainty, but have been less well studied. On the other hand, emissions of NOx should be much better understood; many utility boilers, one of the two dominant sources, have continuous emissions monitors (CEMs, that can also provide SO 2 ), and the studies indicate that for the other dominant source, automobiles, NOx is better estimated than VOCs (e.g., Pierson et al., 1990). Biogenic NOx emissions estimates are still being developed, and may be important in some areas. Given the sensitivity of the air quality models to emissions, these uncertainties appear to be the dominating limitation in our current ability to accurately predict the dynamics of ozone over urban and regional scales. Emissions Processing Methods Emissions processing is a critical step in preparing inputs for air quality models. The emissions model converts the point, area, mobile source emissions to hourly emissions of model species in each grid cell. The conversion procedure is completed in a sequence of steps called temporization, speciation, and gridding. In many cases, emissions models have computational requirements that exceed those requirements for the chemistry transport model. Point sources are defined for emissions processing in terms of their county, process, plant, and stack. Area sources represent emissions of a potential type in a specific region. Mobile sources emissions are based on the Vehicle-Miles-Traveled (VMT) which are converted to inventory pollutant emissions using emissions factors created by running MOBILE5 model. 8

107 The standard procedures for emission processes include the Emission Preprocessor System 2.0 (EPS 2.0) (EPA, 1993), the Flexible Regional Emissions Data System (FREDS) (Modica et al., 1989), and the Emissions Modeling System -95 (EMS-95, formerly GEMAP) (Radian, 1993). These models utilize a lookup procedure in which at each stage a given processor tries to find the properties of each source (e.g. state/county code, process code) in a cross-reference table, which lists the source properties with an associated profile number until a match is found. The profile number is used as an index into a profile table, which provides conversion factors to be used for the source. The lookup process is computationally slow and consists of many redundant steps. MAQSIP uses a fast emissions modeling system; the Sparse Matrix Operating Kernel Emissions (Coats, 1996; Coats and Houyoux, 1996). SMOKE modeling system formulates emissions modeling operations in terms of the mathematics of operator theory. Controls and projection, speciation, and gridding are represented by sparse matrices. The SMOKE computational approach increases computational efficiency enormously. SMOKE uses BEIS-2 biogenic emissions (Pierce et al., 1990) and interacts with MOBILE5 for the calculation of mobile source emissions. A case study using a five day scenario based on the OTAG 1990 inventory (Houyoux et al., 1997) indicates that SMOKE requires 32 times less CPU resources than required by EMS-95. Meteorological Inputs AQMs generally require as inputs the hourly, vertically and horizontally resolved wind fields, as well as hourly temperature, humidity, mixing depth and solar insolation fields. Some AQMs also use the vertical diffusivities, cloud characteristics (liquid water content, droplet size, cloud size, etc.) and rain fall developed from meteorological models. Early studies, as well as many current ones, used objective analysis to interpolate relatively sparse meteorological observations over the modeling domain. Recent AQM applications have found it desirable (because of the sparseness of the data) to use dynamic, or prognostic, meteorological models. Most recently, non-hydrostatic models are being employed. One issue now facing air quality modeling is to assure that the fields developed by the more comprehensive non-hydrostatic meteorological models do not lead to mass inconsistencies when used by the AQMs, and is discussed below. The various types of meteorological models are reviewed by Seaman (this issue), and the discussion here is to highlight the link between the two model types and the effect of various meteorological modeling approaches on air quality model results. For air pollution modeling, the equations governing air motion are generally assumed to be independent of those describing the chemical pollutant dynamics. This is because for problems like smog, it is assumed that the response of pollutant concentrations to emissions controls is so small that they do not significantly impact radiative transfer and cloud formation, and hence weather. Thus, most current air quality models take the emissions and meteorology as inputs derived by techniques that are separate from the air quality model itself. (The emissions preparation techniques do include meteorological effects.) This is the off-line approach (Hansen et al., 1994). When testing control strategies, the meteorological inputs are held fixed to show the response to the emissions changes alone. However, large changes in anthropogenic emissions of some species (e.g., SO 2 ) can lead to increased or decreased aerosol formation impacting cloud formation and light transmission, possibly affecting the meteorology. If such an 9

108 impact were large, the feedback may be non-negligible. However, studies conducted up to this point where a combined model has been used show relatively little impact (e.g., Jacobson, 1997). A drawback of a combined model is that to test emissions strategies, it is necessary to solve for both the meteorological fields (again) as well as how the pollutant concentration fields respond to emissions changes. Given the sensitivity of the meteorological models to inputs, one could even see that there could be an artificial change in the meteorological fields, thus inhibiting the analysis of the control strategy effectiveness. Photolysis Rates Photolysis is one of the most important drivers of tropospheric ozone chemistry. Without photolysis of nitrogen dioxide, there would be no significant production of tropospheric ozone. Likewise, without photolysis of ozone and carbonyls in the troposphere, there would be no production of hydroxyl radicals, and consequently the reactivity of VOCs would be zero. The physical processes which determine the amount of photolysis are: the incidence of sunlight at the top of the atmosphere at a particular zenith angle, molecular absorption of solar radiation by atmospheric constituents, reflection of solar radiation by clouds, aerosols, and by the surface, and scattering of solar radiation by cloud and aerosol droplets and by atmospheric molecules (Madronich, 1987). Although photolysis is critical to determining the reactivities of organic species near the surface, for current determinations of reactivity in urban airsheds (e.g., Carter, 1994), only a relatively simple procedure involving the assumption of clear skies is used to determine photolysis rates. While this may be sufficient to describe the production of ozone during large smog episodes in urban areas, it is insufficient to describe long-range ozone formation over large regions, where clouds invariably affect the photolysis of atmospheric constituents. Even in urban airsheds, the inclusion of clouds can have significant impacts on VOC reactivity, as has been demonstrated by the study of Bergin et al. (1995), which showed significant differences in single-day reactivities of photoreactive species computed using the EKMA box/trajectory model assuming clear skies compared to multi-day reactivities computed by a 3-D Eulerian model assuming cloudy skies. In general, such significant differences in reactivity did not occur for non-photoreactive compounds. Another argument for a more advanced treatment of photolysis in models for reactivity is the significant impact of vertical structure in the overhead ozone column on photolysis in the 290 nm to 320 nm (UV-B) wavelength region, due to multiple scattering (Bruehl and Crutzen, 1989). A number of chemical species apart from ozone have significant molecular absorption in the UV-B region, including several aldehydes and ketones (Martinez et al., 1992) and organic nitrates (Roberts and Fajer, 1989). Sophisticated methods for treating multiple scattering in models, such as the adding method (Lacis and Hansen, 1974) combined with the delta-eddington or similar two-stream techniques (King and Harshvardhan, 1986; Filyushkin et al., 1994) have been known for many years, but the computational demands of such methods have until now discouraged their on-line use in air quality models, which have traditionally employed table look-up procedures. Even 10

109 with the on-line use of advanced methods, there would still be some uncertainty in photolysis rates due to the difficulty of specifying cloud parameters, such as optical depth and scattering asymmetry. An example of a current model which uses look-up tables to compute photolysis rates is MAQSIP. Photolysis rates in MAQSIP are computed prior to a simulation following Chang et al. (1989). In this approach, clear sky actinic fluxes are pre-calculated using a delta-eddington radiative transfer model at specific latitudes and selected hours of the day. A climatological ozone profile is prescribed based on a US standard atmosphere and varies by season and latitude. Aerosol profiles are based on a single profile of aerosol number density (Elterman, 1968), whereas surface albedos are based on values given by Demerjian et al. (1980). The calculated clear-sky actinic fluxes are corrected for cloud effects, and the actual model photolysis rates are then linearly interpolated based on the specific time and grid point location. Future versions of MAQSIP may calculate 4-D photolysis rates by solving the radiative transfer equation for the column atmosphere at each grid point and time step. Gas Phase Chemical Mechanisms Photochemical mechanisms specify the chemical reactions, product yields and kinetics data needed to simulate the photochemical decay of VOC and the production of secondary pollutants. Much of the computational effort in AQMs is required for the numerical solution of the gas phase chemistry, and both computational and memory costs increase rapidly as the number of chemical species increases. As a result, highly simplified condensed mechanisms have been developed for use in AQMs. A number of lumping approaches have been developed to represent the hundreds of VOC found in ambient air using a minimal number of mechanistic entities. One example is the carbon-bond method (Gery et al., 1989) in which functional lumping is used to represent all single carbon bonds in dozens of different VOC as a single mechanistic entity called PAR. Other mechanisms such as SAPRC90 (Carter, 1990) and RADM2 (Stockwell et al., 1990) used molecular lumping so that a single mechanistic entity can represent large classes of similar VOC. The procedures used to convert real world VOC into the mechanistic entities are integral components of the photochemical mechanisms. A consequence of this is that reactivities of the VOC entities simulated in the condensed mechanism may not translate directly to reactivities of real world compounds. This also presents a considerable challenge in comparing VOC reactivities between mechanisms because superficially similar mechanistic entities may have different properties. For example, the toluene entity in different mechanisms might have different reactivities not only due to basic kinetics and product yield differences, but also because the toluene may represent different subsets of real world aromatic compounds. One approach is to augment the condensed mechanisms with explicit chemistry for a particular VOC of interest. This circumvents the complex emissions processing that would be needed to represent the target VOC using the mechanistic entities of the different condensed mechanisms. Even when condensed mechanisms represent explicitly the chemistry of a particular primary VOC, the complex chemistry of subsequently produced intermediates is still 11

110 represented with the highly condensed chemistry using a small set of generic organic intermediates. Thus, the simulation of the reactivity of particular VOCs on multi-day time scales may require more explicit treatment of the intermediate organic chemistry as well. The effect of simplified or explicit intermediate chemistry could be investigated in multi-day trajectory models to compare condensed and explicit mechanisms. Detailed representation of intermediate products is still limited, however, by incomplete knowledge of the products formed. The difficulties inherent in the use of condensed mechanisms could be avoided by using explicit photochemical mechanisms. While fully explicit mechanisms do exist (Derwent et al., 1998; Kerr and Calvert, 1985), there remains considerable uncertainty in the photochemistry of many of the VOC represented in those mechanism. Further, these complex mechanisms are not practical for use in grid models due to their high computational cost. Hales et al. (1993) discussed the use of photochemical mechanisms for reactivity calculations. They listed the following criteria based on mechanism formulation and operational considerations: The mechanism should be designed primarily for applications on the urban scale for oxidant-forming scenarios involving a complicated mixture of VOCs. The mechanism should be consistent with current, consensus kinetics values such as rate constants and product yields. The accuracy of the mechanism should be demonstrated by comparing calculated chemical change with the results of existing smog chamber experiments. The mechanism should not resort to mathematical contrivance to explain basic chemical features. Sufficient documentation should be available for data sources, development process, operational guidance, and perceived limitations and uncertainties. The mechanism should not be so condensed as to ignore significant differences between unlike VOCs. Conversely, the mechanism should not be so large as to be overly expensive or computationally demanding. Clear and complete information should be available to explain conversion of all real world VOC compounds to the reactive entities of the mechanism. The solution algorithm designed to implement the mechanism in a model should be compatible with the above mechanism attributes and capable of representing all necessary chemical detail. The mechanism should be implemented in models capable of performing incremental reactivity calculations, including both grid models and trajectory models. Hales et al. (1993) used these criteria to develop a crude ranking scheme for evaluating photochemical mechanisms. They applied this ranking scheme to the CB4, RADM2 and 12

111 SAPRC90 mechanisms and concluded that SAPRC90 rated barely above adequate, while CB4 and RADM2 were slightly less than adequate for reactivity calculations. Improvements and kinetic updates made to these mechanisms since 1993 may have improved their suitability for reactivity calculations. In addition, several new mechanisms have been developed or are in the process of development. These include the updated SAPRC97, the Regional Acid Chemistry Mechanism (Stockwell et al, 1997), the morphecule mechanism (under development by Harvey Jeffries at UNC) and a new mechanism under development at Environment Canadian (Makar et al., 1998). An important need for the RRWG is to re-asses the criteria proposed by Hales et al. (1993) for evaluating mechanisms, and to conduct detailed reviews of the more recent mechanisms to determine if they are suitable for use in reactivity studies. Mechanism Inter-Comparisons. Comparisons of incremental reactivities using different chemical mechanisms constitute an important component of the uncertainty analysis in the use of reactivity scales (Bergin et al., 1999). Comparisons between mechanisms should be performed in a given model with other model inputs fixed. Such comparisons have been performed in box models (Carter, 1994) and trajectory models (Jeffries and Crouse, 1991). Further, it is necessary to investigate and explain the causes of any differences between mechanisms. Process analysis has been used both to compare mechanisms (Jeffries and Tonnesen, 1994) and to explain differences in incremental reactivities between mechanisms (Jeffries and Crouse, 1991; Bowman and Seinfeld, 1994). As new mechanisms are developed and used in models to simulate reactivity, it is important that comprehensive comparisons should be performed both to explain differences in the mechanisms routine functioning and in their predictions of incremental reactivity. Such comparisons should be performed for a variety of scenario conditions, for a wide range of VOC and NOx emissions levels, and should include comparisons of species concentrations, process diagnostics, and indicator ratios. Clouds and Aqueous Chemistry Cloud an fog droplets can affect gas phase species concentrations by attenuating actinic flux as discussed above, or by scavenging species from the gas phase and mediating the formation of secondary products in aqueous or heterogeneous reactions (Jacob, 1998). For example, NO2 may react on acid/water surfaces to produce HONO. In addition to important effects on PM and water soluble species, aqueous chemistry can affect the gas phase concentrations of O3, and radical species (Lelieveld and Crutzen, 1990). Russell and Dennis (1998) summarize treatment of aqueous chemistry in AQMs: Aqueous phase chemical mechanisms have been implemented in those models where a major focus was acid deposition (e.g., ADOM, RADM, and STEM-II), and are important components of their cloud modules, as well as those following the evolution of aerosols (e.g., Pandis and Seinfeld, 1989; Jacob and Gottlieb, 13

112 1989). Operating multidimensional photochemical models that include aqueous phase chemistry have anywhere from 5 to 100 additional species and 10 to 200 additional reactions. In most cases, the emphasis is on sulfur oxidation routes, and relatively small mechanisms are added. In this case, the added reactions account for the transport of hydrogen peroxide, sulfur dioxide and ozone transport to the droplets, and the resulting sulfur oxidation (e.g., Berkowitz, 1991). More detailed aqueous phase chemical mechanisms exist (see Jacob, [NARSTO]), but they have seen limited implementation in photochemical models due to the tremendous increase in computational requirements (McNair, 1995), and minor impact on model predictions for most gas phase species (Jacob; [NARSTO]). To the extent that heterogeneous chemistry affects radical budgets, it may especially affect the reactivity of VOC species such as HCHO which function primarily as sources of radical initiation. Heterogeneous chemistry is less likely to affect O3 production on the typically sunny, high pressure days most conducive to high O3, but even in that case, night time aqueous chemistry and scavenging may still affect the termination of NOx (Dimitroupolou and Marsh, 1997; Dentener and Crutzen, 1993) and the concentrations of other water soluble species used to evaluate the model. While there is large uncertainty in heterogeneous chemistry, its effects should be considered in uncertainty analysis, and clouds and heterogeneous chemistry must be represented if models are used to simulate particulate incremental reactivities Aerosol Dynamics Routines Deterministically following aerosol dynamics, much like gas phase pollutants are in present models, adds an extra set of physico-chemical processes to simulate. These equations account for the continuous distribution of sizes and composition of the aerosol. Physical processes affecting aerosols include coagulation, evaporation, growth by condensation, formation by nucleation, and deposition by sedimentation. In addition, chemical reactions may take place both in, and on, the aerosol particles. Deterministic simulation of aerosol processes within an air quality model begins with the fundamental equation of aerosol dynamics which describes aerosol transport (term 2), growth (term 3), coagulation (terms 4 and 5), and sedimentation (term 6) (e.g., Friedlander, 1977): n t v I 1 = Un + = β( vv, vnvnv ) ( ) ( vdv ) β( vvnvnvdv, ) ( ) ( ) Cn v Eq. (13) where n is the particle distribution function (which can be a function of time, space, particle size and chemical composition); U is the fluid velocity; I is the droplet current that describes particle growth and nucleation due to gas-to-particle conversion; v is the particle volume; β is the rate of particle coagulation; and C is the sedimentation velocity. One way of modeling the formation and growth of aerosols is done by sectioning the size distribution, n, into discrete size ranges (e.g., Gelbard and Seinfeld, 1980). Then the size and chemical composition of an aerosol is followed as it evolves by condensation, coagulation, sedimentation, and nucleation. Another 14

113 method involves using a functional size distribution, instead of discrete bins (e.g., Whitby et al., 1991). This approach is prompted by the modal structure of the aerosol size distribution. This latter technique is computationally faster, though is less flexible and current implementations do not account for the differences in the thermodynamics and chemistry for aerosols of different sizes. The latter technique has been used in the Regional Particulate Model (RPM) to study aerosol transport and deposition in the United States (Binkowski and Shankar, 1995). Both the discrete and modal approaches require a means to figure out how to grow the aerosol due to gas-to-particle conversion. In turn, that calculation requires determining the partial pressure of the condensing species above the aerosol. While this has been done in reasonable detail for inorganic compounds, the theory for mixtures of organic compounds is under development as explained in the section on aerosol reactivity. Vertical Transport and Diffusion AQMs attempt to simulate atmospheric processes in the troposphere which extends from the ground up to an average altitude of 11 km. The troposphere consists of the planetary boundary layer (PBL) which is defined as the lower levels of the troposphere that are directly affected by the earth's surface and respond to surface forcing (frictional drag, heat transfer, emissions, terrain flow, evaporation and transpiration) with a time scale of about an hour or less, and the free troposphere which is above the PBL. The PBL thickness varies in time and space, ranging from hundreds of meters to a few kilometers (Stull, 1988). Therefore, a successful AQM must simulate vertical motions both in the PBL and the free troposphere. In the real atmosphere, wind can be separated in three categories: mean wind, turbulence, and waves. The transport of heat, momentum, moisture, and pollutants is dominated by the mean wind in the horizontal direction (horizontal advection) and turbulence in the vertical (vertical diffusion). Vertical mean wind velocities are usually much smaller than horizontal wind velocities (2 to 10 ms-1) and are on the order of millimeters to centimeters per second. Waves are frequently observed in the nighttime boundary layer and are generated locally by mean-wind shear of by mean flow over obstacles. The waves transport little pollutants, moisture and heat. Turbulence can be visualized as consisting of irregular swirls of motion called eddies. Usually turbulence consists of many different size eddies superimposed on each other. Much of the boundary layer turbulence is generated by forcing due to surface effects. The size of the largest boundary layer eddies is roughly equal to the boundary layer height and is several orders of magnitude greater than molecular diffusivity. A common approach to study turbulence is to split variables (wind, temperature, etc.) into a mean part and a perturbation part. By applying such a technique to the equations, a number of new terms are created and some of the terms consist of products of perturbation variables and describe the nonlinear interactions between variables. The terms are associated with turbulence in the boundary layer. Unfortunately, after applying the splitting technique, the number of unknowns in the set of equations for turbulent flow is larger than the number of equations. Therefore, the description of turbulence is not closed. 15

114 To treat the closure problem, one approach is to use only a finite number of equations and then approximate the remaining unknowns in terms of known quantities. Two major schools of turbulence closure have appeared in the literature: local and non-local closure. Neither methods are exact and can alter the reactivity of the model significantly. For local closure, an unknown quantity at any point in space is parameterized by values and/or gradients of known quantities at the same adjacent point. Local closure thus assumes that turbulence is analogous to molecular diffusion. For non-local closure, the unknown quantity at one point is parameterized by values of known quantities at many points in space. This assumes that turbulence is a superposition of eddies, each of which transport fluid like an advection process. In Eulerian air quality models, the most widely used local closure method is the K-theory. The concept of K-theory is that turbulence can be simulated in an analogous way to molecular diffusion. It is one of the simplest parameterizations and is used both in the boundary layer and the free troposphere. However, it is valid only over small distances and can be a very poor representation when larger-size eddies are present in the flow. Furthermore, the K-theory can not simulate counter-gradient flows where a turbulent flux flows up the gradient (e.g. from cold to hot). Thus, K-theory is not recommended for use in convective mixed layers (Stull, 1988). The most frequently used non-local closure method is the Asymmetrical Convective Model (Pleim and Chang, 1992). The Asymmetrical Convective Model simulates the effects of convective plumes by mixing material directly from the surface layer to every other layer in the convective boundary layer. Downward mixing however, occurs only between adjacent levels in a cascading manner and is simulated by a local closure method such as eddy diffusion. With this approach, a more realistic simulation of vertical transport within the convective boundary layer is obtained. However, since this method mixes the same amount of mass to every vertical layer in the boundary layer, it has the potential to remove mass much too quickly out of the surface layer. This method also fails to account for the upward mixing in layers higher than the surface layer. This method is applied only to the convective boundary layer. Another non-local closure method is the turbulent transilient theory. This method is based upon the turbulent kinetic energy equation and allows large eddies to transport fluid across finite distances before being mixed with the rest of the environment by the smaller eddies. It provides a framework for considering the ensemble-average effect of many eddies of different sizes on the net non-local mixing in the vertical. Wang (1998) compared three different vertical transport methods: a semi-implicit K-theory scheme (SIK) with local-closure, and asymmetrical convective method (ACM) and turbulent transilient theory (T3) schemes with non-local closure. Of the three schemes, SIK diffused mass most slowly in the vertical direction and maintained the highest amount of ozone precursors in the surface layer. ACM moved mass most rapidly out of the surface layer into other vertical levels. T3 was intermediate between the SIK and the ACM schemes in terms of the rate at which mass was mixed between different layers. O3 is a secondary pollutant and the rate of its formation depends on the amount of available precursors in the system. Even with identical initial condition and input files, different vertical 16

115 diffusion schemes can affect the rate at which emitted precursors are diffused between vertical layers, which in turn affects the rate and efficiency of O3 production in each grid cell. As a result, O3 reactivity in a given cell can differ significantly depending on the scheme used to represent vertical mixing. A grid cell may even change between being NOx-limited or VOClimited by changing the vertical diffusion scheme. The O3 produced and transported in the model will differ spatially and temporally. In some cases, the differences can be as large as twenty to thirty ppb (Wang, 1998). The vertical diffusion scheme interacts non-linearly with other physical and chemical processes in the model, and its impact will change from grid cell to grid cell and from case to case. Much more study is needed to better understand its full impact on model simulations of reactivity based substitution strategies. Chemical Solvers Using operator splitting on the chemistry terms, we can reduce the ADR equation to the form: dci, k = Pi, k LCi, k for i = 1, N (12) dt where we have one ordinary differential equation (ODE) for each species i, and N is the number of species in the chemical mechanism. Subscript k represents each of the grid cells air parcels in the model. P i,j is the rate of production of C i,j found by summing production in all reactions, and L i,j, is the loss frequency of C i,j Equation (1) represents a system of coupled, first-order, nonlinear, ODE s for each cell k. The ODE s are coupled by the terms P i,j and L i,j, which are nonlinear functions of time and C 1 N. We can solve for the species mixing ratios independently in each cell of the grid. Solution of the chemistry term requires approximately a factor of ten more time than the other terms in the continuity equation. Thus, the greatest potential for reducing the computational cost of the chemical tracer model (CTM) is in reducing the number of cells needed to represent the domain and in improving the efficiency of the chemical solver. The time constant of species i, defined as τ i = 1/L i, is a measure of how quickly C i would tend to approach its equilibrium value. The system of coupled ODEs defined by (12) is stiff because the species time constants vary by several orders of magnitude. Systems of stiff ODEs and numerical methods for their solution are discussed in Byrne and Hindmarsh (1987). A system of equations is said to be stiff if the real component of its eigenvalues (λ) are negative, the ratio of the maximum to the minimum eigenvalue is large, and C does not undergo rapid oscillations [McRae et al., 1982]. While the time constant and the eigenvalues are not identical, for many short-lived species τ i λ, so the stiffness of a system can be approximated as: σ=τ max/τ min. A system is mildly stiff for σ 10 2 and stiff for σ > Photochemical mechanisms typically produce systems with σ so these systems are very stiff. Stiff systems are very 17

116 stable, but they are not efficiently solved using traditional ODE solvers because the time step is limited by the species with the fastest time constants. Non-traditional, computationally expensive methods are required to solve stiff systems. The GEAR Solver Gear s method (Gear, 1971) has become the standard against which other methods are compared. The Gear solver is an implicit, backwards difference algorithm in which concentrations from previous time steps are used to predict the concentration at the current time. The algorithm automatically adjusts the size of the time step and the order (the number of previous time steps used) to optimize the solution. The algorithm also estimates the error in the numerical solution at each time step, and the user can specify an error tolerance that constrains the accuracy of the solution. Gear solvers are computationally expensive. Importantly, much of the computation time is required in the first few minutes of a simulation while the order and time step are increasing. If the species sink and source functions are relatively smooth, the integration proceeds very rapidly, typically with t = 10 minutes once a high order has been achieved. Thus, the Gear method works well for box models or trajectory models in which chemistry is solved continuously for a single cell. In a grid models where operator splitting is used, the solver must be restarted at each step of the splitting, and traditional Gear solvers becomes prohibitively expensive due to their high start-up cost. Jacobson and Turco (1994) have developed an optimized matrix algorithm that significantly speeds up the Gear solver for vectorized computers. Although this approach is limited to vectorized supercomputers, it has been implemented in a regional scale AQM (Models3) and can be used for critical applications or as a standard to evaluate other chemical solvers. Saylor and Ford (1996) and Chock et al. (1994) suggest that Gear methods can be competitive in terms of speed and accuracy for global modeling where the advection time step is on the order of one hour (as opposed to about 10 minutes for urban and regional scale AQMs). Sandu et al. (1997) further investigated chemical integrators and found that optimized Gear-type solvers performed best, when considering both accuracy and computational effort, again using chemical integration steps of one hour. This suggests that if one can use large time steps, the advances in optimized Gear-type solvers make that approach very attractive as compared to the hybrid and QSSA solvers currently being used in a variety of models. This may suggest that nested models may benefit from using one type of solver (e.g., Gear-type) for the coarse mesh calculations where the time step may be large, and less accurate, more efficient methods (discussed below) for the fine mesh. Lower Accuracy Methods Accuracy tests of both the chemical kinetics solvers and the transport algorithms indicate that the methods used to integrate the chemistry introduce little error as compared to the advection algorithms (e.g., Odman et al., 1992; Dabdub and Seinfeld, 1994), and it may be 18

117 possible to speed up model execution with a small impact on total numerical error by optimization of the relative amount of error introduced by the advection and chemical integration algorithms. In general, Pi and Li are time varying and depend on concentrations of other compounds. Even though species i might react very rapidly, Pi and Li are rather constant over periods of a minute or so. Thus, analytical solutions to the uncoupled Equation (11) can be used to develop efficient, approximate numerical solutions to Equation (11). Such techniques have been shown to be sufficiently accurate and are an order of magnitude faster than traditional Gear solvers. The two most popular methods for use in photochemical models are the Quasi-steady state approximation (QSSA, Hestveldt et al., 1978) and the Hybrid method (Young and Boris, 1977). Saylor and Ford (1996) and Chock et al. (1994) found that the hybrid method is generally better than QSSA. Mathematical and Computational Implementation There is no known analytical solution to the general ADR equation given in Equation (11). Instead, numerical approaches are used to solved a discretized version. Numerical solution of the complex systems of partial differential equations governing both the evolution of pollutant concentrations requires specialized mathematical techniques, some of them similar to those used in meteorological models. The ADR equation can be dissected into terms describing bulk flow (term 2), turbulent diffusion (term 3) and other processes (e.g., emission sources or chemical reactions, term 4), each having an impact on the time evolution of the transported property (term 1): c t = [ U c+ K c] + [ R( c, t ) + S] Response Convective terms Source Terms In AQMs, the processes have very different time scales (e.g., between species lifetimes, source emissions, and transport). They can be viewed as being relatively independent over a short period of time. This allows the equation to be "split" into separate operators, and is done to greatly shorten solution times (McRae et al., 1982), and improve numerical accuracy. The solution sequence is: c t H = U c+ K c = L ( c) H (13) (14) c t V = w c + z z K c = LV ( c) z (15) 19

118 c t S [ ] = R(c, t) + S = L ( c) S (16) c t Total = ( L + L + L )( c) H V S (17) where LH is the horizontal transport operator, LV is the vertical transport operator and LS is the operator describing other processes, such as chemistry and source emissions. Numerical accuracy is improved and computational times reduced because splitting allows use of solution techniques that are designed to effectively describe specific processes. For example, in a photochemical air quality model, one routine is used for horizontal transport due to bulk winds, and another for vertical motions, which are generally diffusive processes, and a third for the chemistry. Historically, numerical schemes used to calculate the rate of transport have been based on finite difference, then finite element and, more recently, finite volume techniques. Spectral methods have also shown promise, but nonlinear chemical effects tend to degrade solution accuracy in some tests (Hov et al., 1990; Chock, 1985; Odman et al., 1996). Some of the more widely used schemes include: Smolarkiewicz, a finite difference method (Smolarkiewicz, 1983); Bott, a finite difference method (Bott, 1989); chapeau function finite element (McRae et al., 1982); and ASD (Accurate Space Derivative) a pseudo-spectral method (Chock and Winkler, 1994). More recently, Lagrangian approaches have shown promise as well (Chock, 1995; Pudykiewicz et al., 1997). These Lagrangian approaches should not be confused with Lagrangian trajectory models. A problem with numerical solutions of the ADR equations is that the spatial discretization of the modeling region leads to artificial numerical dispersion, which is manifested by the formation of spurious waves and by pollutant peaks being spread out. A number of studies have evaluated the performance of advection solvers to identify their strengths and weaknesses, both alone and in air quality models (Hov et al., 1990; Odman et al., 1996; Chock and Winkler, 1994). As discussed by Byun (1997a&b), meteorological models do not always solve, explicitly, the mass conservation equation, though air quality models usually assume that the three dimensional fields supplied do conserve mass, both globally and locally, using the same discretization methods. If not, the numerical routines in the AQMs will not conserve the mass of the trace species. This has been found to be a problem in the MM5-RADM link. Another problem that can develop is the time discretization of meteorological fields being supplied to the AQM can also lead to mass conservation inconsistencies, or if the various models are not using the same grids interpolation inaccuracies can also lead to problems. In addition to the numerical method being employed, it is important to make sure that there is a consistent treatment of the velocity field between the meteorological model and the air quality model. The advection equation being solved is an expression of mass conservation, and 20

119 is an important requirement in air quality modeling. For an accurate study of source-receptor relationships the models must maintain mass of pollutants emitted into the atmosphere. However, there are several factors that may lead to violation of mass conservation in current air quality modeling practice. The physical law of mass conservation is expressed by the continuity equation. Some meteorological models (e.g., MM5) do not explicitly use the continuity equation in their governing equations set and they diagnose air density (which is a primitive variable in continuity equation) from pressure and temperature using the ideal gas law. If the meteorological model is not conserving energy for some reason (i.e., irreversible effects) this would also affect mass conservation. There are some efforts to include continuity equation in the equations list of meteorological models that are going to be used for air quality modeling (e.g., Byun 1997a) however, this is not going to entirely solve the mass conservation problem as discussed below. The output of meteorological models is stored at a certain interval (e.g., 1 hr) that is usually much larger that the air quality model s time step (e.g., 5 min). It is common to interpolate the stored data and obtain meteorological data for each time step. Since the continuity equation is linear, some people suggest that linear interpolation of density and momentum components would yield a mass conserving set of variables if the originally stored variables (those densities and momenta coming from the meteorological model) satisfied continuity. However, it should be realized that the output from a meteorological model is not the exact solution to the continuity equation, which expresses mass conservation in a continuum. The meteorological model treats the continuum as a discrete set of points at discrete instants separated by a finite time step. Mass conservation in a meteorological model means the satisfaction of some discrete form of the continuity equation. Also, the terms in the continuity equation are not the variables themselves but the rate of change of density and the gradient of momentum. While linear interpolation of the rate of change of density and gradient of momentum would guarantee satisfaction of the discrete conservation equation, interpolation of density and momentum does not. To reconstruct the rate of change and the gradient from the variables requires knowledge of what discretization formulae were used in the meteorological model. In addition, knowledge of the grid (this is not a requirement when the air quality model and the meteorological model use the same grid), time steps during the time interval and some additional data (e.g., the rate of change of a variable is usually expressed as the difference between the values at the current and next time steps divided by the time step and, usually, the value at the next time step is not stored) is required. Meteorological models may use much more complex discrete forms then those used in the air quality models. For example, while explicit single-step methods are preferred in air quality models, stability considerations in meteorological models may favor the use of implicit methods, sometimes with multiple time steps. Mass conservation is guaranteed if the mass continuity equation (Equation (1)) is discretized in the flux-form (or strong conservation form). For reasons explained above, the velocity and density fields output by the meteorological model may not necessarily satisfy the discretized form of the continuity equation as assumed by the AQM. As a solution to this problem, the vertical velocity may be recomputed (e.g., Flatoy, 1994). In this approach, the AQM would interpolate density and horizontal flux components linearly in time and solve Equation (1) for vertical flux and ignore the vertical velocity component coming from the meteorological model. The solution to the species continuity equation (Equation (11) in the 21

120 AQM should also be discretized in the flux form, this time to conserve pollutant mass. The solution technique used for equation (11) should be consistent with the discretization used to satisfy Equation (1). Russell and Lerner (1981) and Easter (1993) describe advection schemes that are consistent with the solution of the continuity equation. The problems noted above suggest that the links between various meteorological models should be understood, and the resulting fields investigated, to assure that no spurious increases (or decreases) in mass occurs during the air quality modeling. UNCERTAINTIES IN MODEL INPUTS: IMPLICATIONS FOR MODEL RELIABILITY It is important to recognize that uncertainties in model components and in model inputs may have a serious impact on model predictions. Given the uncertainty in model simulations, what is the probability that a model would suggest an inappropriate strategy? This question gets at the heart of assessing and comparing emissions control strategies. While we are not yet able to answer this question in detail, significant efforts have been made to quantify the effects on both ozone prediction and control strategy evaluation caused by uncertainties in model components (e.g. the chemical mechanism, wind field generation) and in input data (e.g. emissions rates, wind fields), and from the computational methods applied (e.g. grid size, numerical scheme). Evaluation of model uncertainty presents a challenging problem in part due to the large number of inputs current chemical mechanisms may include hundreds of chemical reactions. Each of these reactions has a number of parameters governing the production and destruction of chemical compounds, including kinetics data and product stoichiometry that vary as a function of sunlight and temperature. While certain reactions and products are known to be more important than others, it is difficult to determine which of these to eliminate in advance from the uncertainty analysis a relatively unimportant reaction with a highly uncertain rate constant may have a similar effect on final model predictions as a crucial reaction with a much better understood rate constant. This complexity presents two crucial steps in evaluating the effects of uncertainty on AQM predictions: Selection of which input parameters and data fields to examine the effect of uncertainty in every parameter and data field characteristic used in a three-dimensional photochemical atmospheric model can not realistically be estimated, so a subset of important parameters must be selected, and Estimation of uncertainty in these identified items parameter and data uncertainties must be estimated by either quantitative analysis such as data with-holding computational experiments or by expert estimation. While, unfortunately, the uncertainty in most inputs and parameters for large atmospheric models are unknown and are not discussed in literature, a few such studies have been published. These include expert reviews of chemical reaction rate constants and product yields (DeMore et al., 1990, 1994, 1997; Stockwell et al., 1994; Atkinson et al., 1989, 1992) and a data withholding 22

121 analysis of wind field uncertainty due to wind speed and direction measurements (e.g., Noblet et al., 1998; Bergin et al., 1999). Sensitivity analysis or uncertainty analysis using less complex models (such as a box model) can be applied in an iterative approach to reduce the set of uncertain parameters for further, more detailed analysis. After the parameter set has been selected, and the uncertainty in each item has been estimated, a number of analysis methods are available for use. The method selected for an uncertainty analysis depends on the number of parameters to be evaluated, the computational requirements of the model, and the detail of information desired. The most common methods used are sensitivity analysis and Monte Carlo analysis. Four types of sensitivity analysis can be defined: Stand-alone tests of the effects of a particular parameterization within a module or mechanism, Response of the model outcome variables to changes in process parameterizations, model structure/numerics, or input variables, Alteration of outcome variable response to an emissions change due to modification of the model or input variables, and Alteration in the emissions control strategy necessary for attainment of an air quality standard. To date, most studies have assessed how changing model inputs would impact predictions of a single outcome variable, O3 concentration (type 1 and type 2). While this is a necessary first step, it is insufficient because outcome variables may not always be accurate measures of system changes. In the classic example, sensitivity simulations showed that biogenic emissions had relatively small effects on peak O3 but had very large effects on the control strategy effectiveness (Chameidies et al., 1988). Thus, a more rigorous uncertainty analysis should also evaluate effects of model input certainty using type 3 and type 4 analyses. For example, Tonnesen (1998) evaluated the effects of uncertainty in an important rate constant on O3 concentration, O3 sensitivity to emissions changes, and changes in O3 attainment strategies. Uncertainty studies can also be extended by considering effects on component model processes in addition to outcome variables (Dennis et al., 1998). The most widely used sensitivity approach is parametric analysis, in which the outputs of a base case are compared to outputs from additional simulations with parameter values changed one at a time (OAT) (Bergin et al., 1998a, 1998b). This approach is valuable when the number of parameters of interest is limited and the computational requirements of the model are very high. Systematic sensitivity analysis techniques, including local techniques such as the direct decoupled method (Dunker, 1984) and Green s function methods (Hwang et al., 1978), and global techniques such as the Fourier amplitude sensitivity test (Koda et al., 1979), represent more efficient means of calculating sensitivities to large numbers of parameters. These systematic methods have been used primarily to study atmospheric chemistry in box model simulations (Falls et al., 1979; Pandis et al., 1989; Milford et al., 1992), but in a few cases have been applied to three-dimensional (3-D) models (Dunker, 1981; Yang et al., 1997). 23

122 This OAT approach may be inadequate in AQMs, however, in which there exist strong non-linear dependencies among the input parameters. Global sensitivity methods such as Monte Carlo techniques are used to propagate probability distributions through the model to estimate uncertainties in output concentrations. Even when using an efficient sampling method such as Latin Hypercube Sampling (Iman and Shortencarier, 1984), Monte Carlo methods can require large sets of simulations to represent uncertainty distributions and can therefore be computationally intensive. This approach has been applied to evaluate descriptions of gas-phase tropospheric chemistry (Thompson et al., 1991; Yang et al., 1995; Gao et al., 1996), uncertainties in a Gaussian plume model (Freeman et al., 1986), and single-cell trajectory models of O 3 formation (Derwent and Hov, 1988). An analysis has been performed using a trajectory version of the CIT AQM (Bergin et al., 1999) to evaluate both secondary pollutant predictions and ozone control strategy evaluation. Preliminary Monte Carlo uncertainty analysis have been performed for three-dimensional photochemical models by Hanna et al. (1998) for a large number of parameters with a sampling size of 50 to study ozone formation, and by Yang et al., (1999) with sample size of 50 to study uncertainties in ozone predictions, control strategy effectiveness and VOC reactivity. The recent AQM and reactivity uncertainty analysis studies are described in Table 1U. After identifying the uncertainties and sources of uncertainty in AQMs, this knowledge should be used to increase reliability in model predictions of air quality and of control strategy efficiency. The most direct method for doing so is to better quantify those parameters identified as contributing most uncertainty. Also, as computational limitations diminish, the more sophisticated statistical approaches can be applied to the complex models to account for interactions between a greater number of parameters. A preliminary study is underway (Bergin and Milford, 1999) that applies a Bayesian technique to the results of a previous Monte Carlo study in order to incorporate pollutant measurements into model predictions for a more reliable estimate of uncertainty. Techniques such as this may allow better characterization of uncertainty as well as increasing (or decreasing) confidence in model predictions. DIAGNOSTIC METHODS Brute Force Sensitivity The simplest and most common approach to sensitivity studies is a brute force method in which a base case simulation is performed, and then the simulation is repeated using a change in some model input. A common sensitivity study is to investigate O 3 sensitivity to precursor emissions by, for example, reducing VOC or NO x emissions by 50% or some other level representative of a control scenario. Incremental reactivity is defined as the derivative of O 3 with respect to precursor emissions and this can be approximated in a brute force sensitivity study by using a small, finite change in precursor emissions and calculating the difference in ozone concentration ([O 3 ]) between two simulations: 24

123 d[o 3 ] [O 3] [O 3] new [O3] base = (18) dei Ei Ei, new Ei, base where E j represents base case emissions of some precursor j, and E j,new represents simulation with emissions increased or decreased by some small amount. The accuracy and stability of results calculated with Eq 4.1 vary with the choice of chemical solver and user specified error tolerance levels. To date, the primary concern in evaluating the accuracy of chemical solvers has been the accuracy with which they predict species concentrations, particularly O 3. Numerical errors of less than ±1% are generally acceptable because uncertainty in measured O 3 is much larger, and because such small errors have little effect on conclusions drawn from typical modeling studies. When models are used to estimate incremental reactivity, however, numerical error can be large relative to the small changes in O 3 resulting from small emissions changes. Thus, error tolerances that were acceptable for O 3 can become unacceptably larger for small increments in emissions and [O 3 ]. To illustrate this, Figure 1(top) shows model predicted [O 3 ] for three different Gear error tolerances in simulations where NO x emissions were incremented in a range from 2% to +2%. An error tolerance of 3x10-3 (which is the default EKMA setting) produces errors of less than ±0.2 ppb relative to the most stringent error tolerance of Figure 1(bottom) shows that these small errors in [O 3 ] translate to large errors in [O 3 ]/ E NOx Thus, care must be exercised in simulating incremental reactivity to ensure that the emissions increment is large enough so that numerical error does not overwhelm the O 3 response. If the emissions increment is too large, however, it does not provide a good approximation of the derivative because the O 3 response is non-linear with the size of the emissions increment. Figure 2 illustrates this for formaldehyde (HCHO) incremental reactivity where HCHO incremental reactivity can vary between 0.34 and 0.53 depending on the size of the emissions increment. A more accurate estimate of incremental reactivity can be calculated using the centered-difference with positive and negative changes in precursor emissions: d[o ] dei Ei 3 [O 3] [O 3] + 2% [O3] 2% = E i, + 2 E i, 2% (19) where, for example, precursor emissions are incremented by ±2%. Using Eq 4.2 the brute force method can provide accurate estimates of incremental reactivity for [O 3 ] values as low as a ±0.1 ppb using a Gear solver with a strict error tolerance in a trajectory model, using single precision variables. In grid models, greater numerical error is introduced due to low order accuracy chemical solvers and numerical error in transport solvers. Tonnesen et al. (1998) found numerical error of less than ±0.5 ppb in brute force sensitivity runs in a grid model. Careful study is required to determine the size of emissions increments which should be used and the accuracy with which incremental reactivity can be calculated in particular grid models. An advantage of the brute-force approach is its simplicity of implementation. It is also useful for sensitivity studies designed to evaluate the effects for large changes in precursor 25

124 emissions. In these cases it is necessary to consider the non-linear dependence of the O 3 response on the size of the emissions change. The brute force method cannot, however, be used to determine sensitivity to small emissions changes for which the O 3 response is overwhelmed by numerical error. A highly accurate and efficient method for calculating incremental reactivity for very small increments (thereby avoiding the limitations in the brute force approach) is discussed next. Direct Sensitivity Analysis Direct sensitivity analysis is another technique to better understand model response. In this case, sensitivity parameters are found as the partial derivative of the predicted concentrations to specified model inputs or parameters: S ij c cp ( P cp i j + j) ( j) = P P j j (20) where S ij is the sensitivity coefficient of the concentration of the i th species to the j th parameter, P j is the j th model parameter (or input) and P j is a small perturbation in the parameter. This can be found in either via a brute force approach as described above, or a more formal method. Brute force is conceptually the simplest but has several significant limitations. First, if the number of sensitivity coefficients desired is large, this becomes overly burdensome and computationally inefficient. Second, given the numerical noise in model calculation (e.g., due to the transport algorithms), if the parameter is not perturbed significantly, it can be difficult to accurately calculate the model response. Formal sensitivity techniques, such as the Decoupled Direct Method (DDM) and the Adjoint approach, can alleviate some of the limitations. Application of DDM in air quality modeling was pioneered by Dunker (1980, 1984). Yang et al., (1997) more recently implemented it in the CIT model to show how the model responded to various emissions and a number of model parameters. It has proven to be a computationally effective method to find model sensitivities, perform uncertainty analysis, conduct source apportionment and quantify VOC reactivities. DDM is particularly powerful for conducting reactivity analysis. The sensitivity coefficient of O 3 to the emissions of a particular species is that species reactivity. Thus, a single simulation where, say, 20 sensitivity coefficients of ozone to emissions of different VOCs are found, provides, directly, the 20 VOC species reactivities. Another advantage of this method over a brute force approach for this type of calculation is that it reduces the numerical noise. Studies using DDM to quantify reactivity include Yang et al. (1996) using a box model and Yang et al., (1998) and Kahn et al. (1998) using three-dimensional airshed models. Because of its computational efficiency, use of DDM makes it possible to conduct reactivity analysis at regional scales (e.g., Kahn et al., 1998). These studies found that using DDM provided very similar results as compared to the brute force approach, and were computationally more efficient. As discussed elsewhere, DDM has been used in a Monte Carlo analysis of reactivity uncertainty (Kahn et al., 1998). 26

125 Process Analysis Process analysis is a technique for explaining a model simulation it stores the integrated rates of species changes due to individual chemical reactions and other sink and source processes (Jeffries and Tonnesen, 1994). AQMs typically operate by using the species rates of changes to predict the temporal evolution of species concentrations, but information regarding the particular reactions and processes that cause that evolution are not saved or evaluated. By integrating these rates over time and outputting them at hourly intervals, process analysis provides valuable diagnostics that can be used to explain a model simulation in terms of the budgets of radical initiation, propagation and termination, production and loss of odd oxygen and O 3, and conversion of NO x to inert forms, as well as the effects of transport and other sink and source terms. While process analysis can provide an improved understanding of cause and effect in a model simulation, it does not necessarily have predictive ability, i.e., it does not predict the reactivity of VOCs. Process analysis can, however, be used to explain reactivity by explaining how O 3 was produced in the base case scenario and explaining exactly what changed in the sensitivity simulation. Similarly, incremental reactivity has predictive power, but it does not necessarily have explanatory power it can be difficult to interpret on the basis of sensitivity simulations alone why a VOC exhibits a particular reactivity. Thus, incremental reactivity and process analysis are complementary approaches that can be combined to more fully characterize the system. For example, process analysis has been used to explain the incremental reactivity of a given VOC by explaining its effects on radical initiation, NO x termination, and the changes in the amount of other VOC that react (Jeffries and Crouse, 1991; Bowman and Seinfeld, 1994). Process analysis has been implemented in grid models (Jang et al, 1995). In addition, algorithms have been developed to track mass transport in grid models to explain how emissions in a given cell affect species downwind (Lo and Jeffries,1997). A limitation of these approaches in grid models is that they can require large resources for storage and post-processing of integrated reaction and process rates. These limitations have been circumvented by performing processing of the integrated rates within the model and outputting a limited set of process diagnostics (Tonnesen and Dennis, 1998a) or by providing a flexible user interface to select particular process diagnostics as in Models3/CMAQ. These diagnostics are then immediately available to be analyzed in conjunction with species concentrations fields. It remains difficult, however, to fully characterize the chemical budgets of particular precursor emissions sources in grid models because the Eulerian grid structure is not conducive to evaluating the transformations in air parcels being transported across the grid. A tracer approach that addresses this problem is described in the next section. 27

126 Ozone Source Apportionment Technology Source-receptor analysis is fairly easy to study in Lagrangian models, although the simplistic Lagrangian treatment of transport raises uncertainty about the results of these studies. Source-receptor relationships are much more difficult to study in Eulerian models because it is computationally expensive to maintain a record of the mass transport between cells. Eulerian models typically use the calculated rate of mass flux to update the concentration in each cell but do not compute or output the actual mass fluxes between grid cells. The CAMx model (ENVIRON, 1997) employs a system of tracers to determine the emissions source regions which affect O 3 production throughout the model domain. OSAT also maintains a record of whether O 3 production occurred under NO x - or VOC-limited conditions (Yarwood et al., 1996). OSAT appears to be a combination of the sensitivity approach and the process analysis approach, and it is intended to provide guidance in identifying effective precursor control measures. Mass tracking and tracer analyses are potentially useful in model studies of reactivity because: (1) they further elucidate at a process level the factors which contribute to high O 3. (2) they provide detailed information regarding source-receptor relationship. Continued development and application of these diagnostic tools will be useful in model studies of reactivity. Indicator Ratios Ratios of particular combinations of species have been proposed to be useful as indicators of O 3 sensitivity to changes in precursor emissions (Milford et al. 1994; Sillman 1995, 1998). Several indicators have been proposed using a steady state analysis of radical budgets (Sillman, 1995; Kleinman, 1994; Kleinman et al., 1998). Tonnesen and Dennis (1998a&b) have distinguished between indicators of local odd oxygen production (P(O x )) and O 3 concentration ([O 3 ]) sensitivity and have analyzed indicators in terms of radical propagation efficiency. Following on the work of Johnson (1984), Blanchard et al. (1993) and Chang et al. (1997) have developed an indicator based the extent to which the available NO x has been utilized to achieve maximum potential [O 3 ]. In theory, indicators could be used to estimate P(O x ) and [O 3 ] sensitivity to precursor emissions independently of models. Tonnesen and Dennis (1998b), however, found in modeling studies that indicators of [O 3 ] sensitivity successfully distinguished the extreme NO x -limited and VOC-limited conditions, but were less useful for distinguishing less extreme conditions. Lu and Chang (1998) found that values of the indicator of [O 3 ] sensitivity calculated in a model in the San Joaquin Valley differed from the values reported in other studies. Nevertheless, indicators should still provide a powerful diagnostic test to evaluate models. Sillman et al.(1995, 1997a&b) have used measurements of indicators to evaluate model applications and to identify model scenarios that failed to match observations. Local indicators of P(O x ) sensitivity (Tonnesen and Dennis, 1998a) may better discriminate between NO x - and VOC-limited conditions than do the long-lived indicators. Measurements of these indicators could be useful in reactivity substitution 28

127 strategies as a means for determining whether particular regions are NO x -limited or VOClimited, or at conditions of MIR, MOR, or some other state. To date, the usefulness of indicators for predicting O 3 sensitivity has only been simulated in model sensitivity runs. Before the indicators concept or particular values of indicator ratios can be used with confidence, this predictive ability must be experimentally tested in controlled experiments. Such experimental testing could be conducted in environmental chambers or in other well defined conditions in which P(O x ) and [O 3 ] response to emissions changes can be carefully measured. TYPES OF MODELS Trajectory and Box Models In trajectory models the continuity equation is solved using the wind field as the reference frame. In box models the advection and dispersion terms are considered to be negligible. In practical terms these two approaches are quite similar as each reduces the ADR equation to the simpler form of Equation (12). Because of their simplicity and efficiency, trajectory models have several advantages relative to grid model: They are easier to code and operate compared to grid models; It is easy to specify model inputs and to vary inputs in sensitivity simulations; Large numbers of simulations can be rapidly performed for sensitivity studies. They can include large, explicit chemical mechanisms which would be prohibitively expensive in grid models. Easy to modify mechanisms to include explicit chemistry for particular VOCs. Trajectory models can be used to evaluate chemical effects in isolation from meteorology The Lagrangian reference frame facilitates evaluation of source-receptor relationships. Their numerical efficiency allows for simulation of climatological outcomes (i.e., simulations over seasons or years). Disadvantages of trajectory models include the following: They have overly simplistic representation of transport and turbulent dispersion potentially causing them to underestimate dispersion and overestimate species concentrations. 29

128 Difficult to represent multi-day scenarios, especially those with complex mixing including stagnation, re-circulation and re-entrainment of polluted air from above the PBL. They provide poor temporal-spatial resolution output is defined only at a limited set of points and times along the parcel trajectory. This makes it difficult to evaluate reactivity effects integrated over a geographical area or integrated over time at a given site. To address this last deficiency, trajectory models have been defined using multiple air parcels moving through a fixed grid with the ensemble average of parcels within grid cells used to calculate species concentrations on the grid (Simpson et al., 1992). These approaches have also been useful in global scale chemical tracer models (CTM) for which it is still impractical to solve complex photochemistry on an adequately resolved grid (Derwent, 1998). Multiple-parcel trajectory models are also useful in cases where there is particular concern for attributing species to their source regions (Simpson, 1997). For example, the European EMEP modeling program is particularly concerned with creating blame matrices to determine which sources contribute to high pollutant levels. Here, we describe several trajectory or box models that have been used in reactivity studies. OZIP/EKMA There exist several versions of the U.S. EPA's OZIPP (Ozone Isopleth Plotting Package) computer modeling program (e.g., OZIPM-4, Hogo and Gery, 1988). The OZIPP programs employ a trajectory-based air quality simulation model, in conjunction with the Empirical Kinetics Modeling Approach (EKMA), to relate ozone concentrations to levels of VOC and NO x emissions. Early OZIPP versions were designed to provide a rigid structure within which State Implementation Plans (SIPs) could be formulated. OZIPR (Gery and Crouse, 1990) is a researchoriented version that was created to provide a more flexible and functional research tool. Its structure is similar to OZIPM-4, however, so the published user's guide (Hogo and Gery, 1988) for that program serves as a reference for the development and inherent structure of OZIPR. OZIPR simulates complex chemical and physical processes of the lower atmosphere through use of a trajectory model. The physical representation is a well-mixed column of air extending from the ground to the top of the mixed layer. This idealized air column moves with the wind (along the wind trajectory), but cannot expand horizontally. Emissions from the surface are included as the air column passes over different emission sources, and air from above the column is mixed in as the planetary boundary layer (PBL) height rises during the day. OZIPR provides a flexible interface that can read complex chemical mechanisms, initial and boundary conditions, and diurnally varying times series of precursor emissions, temperature, humidity, and deposition velocities. It also allows default or user defined specifications of the temporal evolution of the planetary boundary layer height. A Gear method is used to solve the photochemistry. Emissions and deposition are represented as first-order reaction rates and are solved simultaneously with the chemistry. 30

129 An important limitation of OZIPR is that it does not simulate dilution of the species in the box caused by horizontal advection and dispersion. In addition, multiple day simulations are difficult to perform, and it does not represent the chemistry and transport of trace species above the mixed layer. A key feature of OZIPR is that it is automated to perform a large set of simulations at specified intervals of VOC and NO x precursor levels. This facilitates the creation of response surface plots showing model output variables as a function of initial precursors levels. The most familiar such response surface plot is the O 3 isopleth diagram (U.S. EPA, 1983), but OZIPR is also automated to create response surfaces for other model species, for a large set of process diagnostics, and for the incremental reactivity of NO x, the VOC mixture, and the individual VOC species. OZIPR also outputs integrated reaction rates for the full mechanism, and an associated post-processing program can be used to perform a process analysis that describes the contribution of individual VOC to important process rates such as radical initiation, NO x termination and O 3 production (Jeffries and Tonnesen, 1994; Tonnesen and Jeffries, 1994). The Harwell Trajectory Model The Harwell model is a single layer trajectory with an aloft layer in the free troposphere (Derwent et al., 1998). It is a regional scale model that uses 300 back-track air mass trajectories each of 96 hours duration, arriving at a grid of 60 arrival points across the British Isles on 5 ozone episode days during 1993 and Features of the model are listed below: diurnal variations in boundary layer depth, temperature and humidity are fully treated. 120 emitted VOCs are treated from 36 emission source categories and emissions of SO2, NOx, CO, methane and isoprene from 11 source categories on a 10km x 10km emissions grid over the United Kingdom and 50km x 50km over the rest of Europe. the model uses a Master Chemical Mechanism which contains 2410 chemical species and 7100 chemical reactions which can be viewed at: the inorganic chemical reactions of the simple atoms and radicals containing oxygen, hydrogen, nitrogen and sulfur and those of CO, employ evaluated rate coefficients with their full temperature and pressure dependencies from the IUPAC and JPL evaluations. the 1300 photolysis reactions of the photochemically-labile species are given a full timeof-day dependence using a two-stream sixth-order code which includes treatment of the stratospheric ozone amount, surface albedo and the background aerosol. the model contains a treatment of the dry deposition of O3, SO2, HNO3, H2O2 and 101 peroxyacetylnitrate analogues. the system of stiff differential equations describing the chemical development of the 2510 species is integrated with a Gear s method integrator with no assumption of photochemical steady states (even for O1D) with automatic accuracy control to 0.1%. 31

130 AES Box Model Ongoing research at the Atmospheric Environment Service (P. Makar, personal communication, 1998) has combined a gas-phase tropospheric reaction mechanism, a 1-D radiative transfer model, and a standard Gear-type and the Canadian National Air Pollution Surveillance (NAPS) measurement data to determine ozone sensitivities to initial NO x and VOC mixing ratios. Measurement records at 15 sites across Canada were extracted from the database and were used to generate one-hour ozone forecasts. The initial mixing ratios of NO x and anthropogenic VOCs were increased and decreased by 25% to determine the effect of changes in the ambient concentrations of these variables on local ozone production. VOCs were varied individually and as a group, to determine the net VOC effect and the effect of individual classes of VOCs on ozone. The results of these simulations were used to construct finite difference sensitivities (O 3 /sensitivity variable), and were expressed as (ppbv change in O 3 per % change in variable), to allow local effect comparisons between sites. Results showed both differences and similarities across regions. Industrial Ontario areas had negative sensitivities to the initial NO concentration, indicating that decreases in NO would lead to increases in O 3 (due to reduced NOx titration), while Greater Vancouver sites had minimal NO sensitivity. Ontario sites had positive sensitivities for net VOCs; reductions would lead to decreases in O 3 ; Vancouver again had a smaller effect. Site-by-site examination of individual VOC sensitivities showed that the greatest sensitivity of VOC classes was for internally bonded alkenes at most sites. The next most reactive classes varied from site to site, with either end-bond alkenes or higher aromatics having the next most important effect after internal alkenes. Magnitudes of the sensitivities showed a large variation of effects even within the same region or industrial area (Figure 3 shows sensitivities for model VOCs and NO at five Ontario sites). The study s results to date indicate that region-and VOC-class-specific VOC reductions may have the greatest impact on ozone concentrations. The study is currently being revised to include sensitivities towards biogenic hydrocarbon mixing ratios. EMEP Trajectory Model The EMEP photochemical trajectory model provides another example, in addition to OZIP/EKMA, of a trajectory model with a long history of regulatory and policy use. The EMEP model has been used extensively in Europe to develop source-receptor relationships for O3, acid deposition, and eutrophication (Simpson et al., 1995a&b). Extensive model scenarios have been run to develop seasonal or climatological averages (Simpson, 1992, 1997). While the EMEP model has the same limitations of other trajectory model, it provides a means to evaluate two areas that may be of interest in studies of reactivity: source-receptor effects and climatological effects. Urban/Regional/Multi-Scale Grid Models Eulerian models solve Equation (11) for a gridded domain using operator splitting as described above. Advantages of grid models include the following: 32

131 They contain detailed descriptions of meteorological and transport processes. Model predicted species concentrations are defined over the full geographical and temporal domain. Designed to simulate multi-day scenarios. Disadvantages of grid models: Expensive to maintain and run. Large inputs and output data sets make it difficult to perform and evaluate large numbers of sensitivity simulations. Difficult to determine cause and effect because of the complexity of interaction between model components. Limited to condensed chemical mechanisms which are typically hard-coded and difficult to change. Table 1 (from Russell and Dennis, 1998) summarizes aspects of some of the Eulerian AQMs that are in current or recent use. Due to the large number of grid models that have been used, we will not provide detailed comments on individual models, although we will comment on several noteworthy models that were not included in Table 1. A new regional scale chemical tracer model (CHRONOS) has been developed at the Canadian Atmospheric Environment Service (Pudykiewicz et al., 1997). While this model uses an innovative semi-lagrangian advection solver, it also employs the somewhat dated ADOM chemical mechanism. Environ Corporation has developed an multiscale Comprehensive Air Quality Model with extensions (CAMx) (Reynolds and Roth, 1997) that uses an extensive system of tracers (Yarwood et al., 1996) that was described above. The Urban-to-Regional Multiscale Model (which is listed in Table 1) is noteworthy because it allows for an irregular, multiscale, spatially varying grid resolution, similar to the approach used in computational fluid dynamics. Unlike nesting, the transport algorithm allows for arbitrary grid structures, and can minimize the number of computational nodes needed to have a fine grid over a non-rectangular region (e.g., the coastline of the eastern United States). The Urban-to-Regional Multiscale has been applied to Southern California (Odman et al., 1994), the Northeastern United States (Kumar and Russell, 1996) and the eastern United States, covering a domain similar to the OTAG region. URM has been used to conduct control strategy analyses for ozone, aerosols and acid deposition, and using the direct differential method (DDM, described below), has been used for reactivity assessment. The Multiscale Air Quality Simulation Platform (MAQSIP ) is the air quality modeling framework within the Environmental Decision Support System (EDSS) which was developed as a prototype for the EPA Models-3 system. MAQSIP provides a base for the next generation, modular Eulerian air quality models. The modular design and flexibility utilized in MAQSIP 33

132 allow it to retain its state-of -the science over time with future development and implementation of new science modules as appropriate. The MAQSIP has been driven by meteorological inputs provided by the Penn State/NCAR Mesoscale Model (MM5) (Grell et al., 1994). The emission model SMOKE provides processing of emission inventories from different sources including the NAPAP emissions. MAQSIP applications include 120 day simulation developed for the Seasonal Model for Regional Air Quality (SMRAQ) performed for the SouthEastern States Air Resources Managers (SESARM). MAQSIP configuration for the SMRAQ seasonal runs uses a modified version of the Carbon Bond IV (Gery et al., 1989) chemical mechanism, a flux-form advection scheme (Bott, 1989), a K-theory scheme for turbulent vertical redistribution of pollutants (Alapaty et al., 1997), and a dry deposition scheme (Walcek et al., 1986). Clear-sky photolysis rates are calculated following Madronich (1987). The cloud package include the deep convective parameterization scheme employed in the MM5 (Kain and Frisch, 1993). Global Models Concerns over tropospheric ozone have expanded within the last decade to include not only the prevention of intense smog episodes in urban areas, but also the alleviation of chronic human exposure at lower ozone doses, the protection of crops, and the mitigation of global warming. As ozone standards fall closer to global background values in response to these concerns, there is a corresponding rise in the need to model the formation and transport of tropospheric ozone throughout the entire world. The global modeling of tropospheric ozone requires spatial and temporal resolution that can best be achieved only with three-dimensional chemical transport models (CTMs). Although two-dimensional, zonally (i.e., longitudinally) averaged CTMs with realistic photochemical schemes have been have been used to assess global tropospheric ozone impacts (e.g., Isaksen and Hov, 1987; Hough and Derwent, 1990), they fail to account for continent-ocean differences in emissions, for zonal transport, and for the nonlinear dependence of ozone production on NOx (Kanakidou and Crutzen, 1993). On the other hand, 3-D CTMs are computationally intensive, and require less detailed treatment of chemistry than in many urban and regional models in order to produce tractable simulations (Houweling et al., 1998). Moreover, their lack of resolution on regional and urban scales limits their ability to simulate the impacts of intense pollution plumes associated with concentrated industrial sources, which may represent an important nonlinear chemical processing of NOx emissions (Sillman et al., 1990). An important issue in the design of 3-D global CTMs is the temporal resolution of the input meteorological fields. In order to conserve computational resources, some CTMs obtain these fields as monthly averaged data from general circulation models (GCMs) used in climate studies. An intercomparison of models by the IPCC (1994) revealed, however, that the use of monthly averaged circulations significantly distorts the transport of global tracers compared to the use of daily varying circulations. More recently, 3-D CTMs have been developed which either use more temporally-resolved GCM output (Wang et al., 1998), or combine the predictive capability of a GCM with a 3-D CTM, resulting in a hybrid chemistry-climate model. (e.g., Roelofs et al., 1997). The hybrid models can not only avoid the inaccuracies associated with lack of temporal resolution in meteorology, but can also simulate feedbacks involving radiative- 34

133 chemical-dynamical coupling, a potentially important issue in climate studies. Their formulation based on the primitive equations of dynamic meteorology, however, can result in an overly large computational burden. An alternative to primitive equation hybrid models is provided by global balance chemistry-climate models (Tie et al., 1991; Olaguer, 1998), which employ a lower order version of the hydrodynamic equations to lessen computational demand. 3-D global CTMs or chemistry-climate models can be useful in the context of reactivity studies as a means of specifying boundary conditions for regional models, or for determining very long range reactivities related to background ozone. They can also be used to study global collateral impacts, such as persistence, global warming, and stratospheric ozone depletion, especially for short-lived compounds which degrade into longer-lived intermediates. Their usefulness would be tremendously improved if regional models were nested into them in limited areas of the globe, where intense anthropogenic emissions of ozone precursors warrant more spatial resolution. REFERENCES Arnold J. R., Dennis R. L. and Tonnesen, G. S. (1998): Advanced techniques for evaluating Eulerian air quality models: background and methodology. In: Preprints of the 10th Joint Conference on the Applications of Air Pollution Meteorology with the Air & Waste Management Association, January 11-16, 1998, Phoenix, Arizona. American Meteorological Society, Boston, Massachusetts, paper no. 1.1, pp Atkinson, R.; Baulch, D.L.; Cox, R.A.; Hampson, R.F. Jr.; Kerr, J.A.; Troe, J. (1989): J. Phys. Chem. Ref. Data, 13, Atkinson, R.; Baulch, D.L.; Cox, R.A.; Hampson, R.F.; Kerr, J.A.; Troe, J. (1992): J. Phys. Chem. Ref. Data, 21, Baugues, K. (1990). Preliminary Planning Information for Updating the Ozone Regulatory Impact Analysis Version of EKMA, Draft Document, Source Receptor Analysis Branch, Technical Support Division, U. S. Environmental Protection Agency, Research Triangle Park, NC (January). Bergin, M.S.; Milford, J.B. (1999): Application of Bayesian Monte Carlo analysis to Lagrangian Photochemical Air Quality Modeling, manuscript to be submitted.. Bergin, M.S.; Noblet, G.S.; Petrini, K.; Dhieux, J.R.; Milford, J.B.; Harley, R.A. (1999): Formal Uncertainty Analysis of a Lagrangian Air Pollution Model, Environ. Sci. & Technol., in press. Bergin, M.S., Russell, A.G., Carter, W.P.L., Croes, B., and Seinfeld, J.H. (1998a): "VOC Reactivity and Urban Ozone Control." Encyclopedia of Environmental Analysis and Remediation, pp J. Wiley & Sons, New York, NY. Bergin, M.S.; Russell, A.G.; Milford, J.B. (1998b): Environ. Sci. Technol, 32,

134 Bergin, M.S., A.G. Russell, and J.B. Milford (1995): Quantification of individual VOC reactivity using a chemically detailed, three-dimensional photochemical model, Environ. Sci. Technol., 29, Binkowski F. S. and Shankar U. (1995): The Regional Particulate Model. 1. Model Description and Preliminary Results. Journal of Geophysical Research 100, 26,191-26,209. Blanchard, C. L., P. M. Roth, and H. E. Jeffries (1993): Continuing development of a methodology for developing for assessing preferences for reductions in VOC and NOx emissions reductions in non-attainment areas, Air and Waste Management Association Conference on Regional Photochemical Measurement and Modeling Studies, San Diego, CA, November Bott, A., (1989): A positive definite advection scheme obtained by nonlinear renormalization of the advective fluxes, Mon. Wea. Rev., 117, Bowman, F. M. and Seinfeld, J.H. (1994b). Atmos. Environ., 28: Bruehl, C. and P.J. Crutzen, (1989): On the disproportionate role of tropospheric ozone as a filter against solar UV-B radiation, Geophys. Res. Lett., 16, Byrne G. D. and A. C. Hindmarsh (1987): Stiff ODE Solvers: A Review of Current and Coming Attractions, J. Comp. Phys., 70, Byun, D. W. (1998a): Dynamically consistent formulations in meteorological and air quality models for multi-scale atmospheric studies: I. Governing equations in a generalized coordinate system. Submitted to Journal of Atmospheric Science. Byun, D. W. (1998b) Dynamically consistent formulations in meteorological and air quality models for multi-scale atmospheric studies: I. Mass conservation issues. Submitted to Journal of Atmospheric Science. Carter, W.P.L. (1990): Atmos. Environ, 24A, Carter, W.P.L. (1994): Development of ozone reactivity scales for volatile organic compounds, Air and Waste, 44, Carter, W.P.L. (1995): Atmos. Environ., 29, Carter, W.P.L. (1994): Report to Auto/Oil Air Quality Improvement Research Program, March.. Chameides, W.L., R. Linsay, R. Richardson, and C. Kiang (1988): The role of biogenic hydrocarbons in photochemical smog: Atlanta as a case study, Science 241, Chang, J. S, R. A. Brost, I. S. A. Isaksen, P. Middleton, W. R. Stockwell, and C. J. Walcek, A (1987): Three-dimensional Eulerian Acid Deposition model: Physical concepts and formulation, J. Geophys. Res., 92, 14,681-14,

135 Chang, T. Y., D. P. Chock, B. I. Nance, and S. L. Winkler (1997): A photochemical extent parameter to aid ozone air quality management, Atmos. Environ., 31, Chock D. P. and Winkler S. L. (1994): A Comparison of Advection Algorithms Coupled With Chemistry. Atmospheric Environment 28, Chock D. P. (1985): A comparison of numerical methods for solving the advection equation, II. Atmospheric Environment 19, Coats, C.J., Jr (1996): High performance algorithms in the Sparse Marix Operator Kernel Emissions modeling system. Proceedings of the 9th joint conference on applications air pollution meteorology of the AMS and A&WMA, Atlanta.. Coats, C.J., and M. Houyoux (1996): Fast emissions modeling with the Sparse Matrix operator kernel Emissions modeling system. The Emissions Inventory: Key to Planning, Permits, Compliance, and Reporting, A&WMA, New Orleans, LA. Dabdub D. and Seinfeld J. H. (1994): Numerical advective schemes used in air quality models sequential and parallel implementation. Atmospheric Environment 28, Demerjian, K.L., K.L. Schere, and J.T. Peterson (1980): In Advances in Environmental Science and Technology, Pitts, J.N., R.L. Metcalf, and D. Grosjean (Eds.), John Wiley & Sons Inc., vol 10, pp , DeMore, W.B.; Golden, D.M.; Hampson, R.F.; Howard, C.J.; Ravishankara, A.R.; Kolb, M.J.; Molina, M.J. (1997): Chemical Kinetics and Photochemical Data for Use in Stratospheric Modeling, Evaluation No. 12, JPL Publication No. 97-4, Jet Propulsion Laboratory, Pasadena, CA. DeMore, W.B.; Sander, S.P.; Golden, D.M.; Hampson, R.F.; Kurylo, M.J.; Howard, C.J.; Ravishankara, A.R.; Kolb, M.J.; Molina, M.J. (1994): Chemical Kinetics and Photochemical Data for Use in Stratospheric Modeling, Evaluation No. 11; JPL Publication No , Jet Propulsion Laboratory, Pasadena, CA. DeMore, W.B.; Sander, S.P.; Golden, D.M.; Molina, M.J.; Hampson, R.F.; Kurylo, M.J.; Howard, C.J.; Ravishankara, A.R. (1990): Chemical Kinetics Data For Use In Stratospheric Modeling; Evaluation No. 9; Jet Propulsion Laboratory, Pasadena, CA. Dimitroupolou, C. and A. R. W. Marsh (1997): Modeling studies of NO3 nighttime chemistry and its effects on subsequent O3 formation, Atmos. Environment, 31, Dennis, R. L., J. R. Arnold, G. S. Tonnesen, and Yonhong Li, (1998): A new response surface approach for interpreting Eulerian air quality models sensitivities, Computer Physics Communication, in press. Dentener, F. J. and P. J. Crutzen (1993): Reaction of N2O5 on Tropospheric aerosols: Impact on global distributuions of NOx, O3 and OH., J. Geophys. Res., 98,

136 Derwent et al., (1998): Photochemical ozone creation potentials for organic compounds in northwest Europe calculated with a Master Chemical Mechanism, Atmospheric Environment, 32, Derwent, R. G. and Jenkin, M. E. (1991): Hydrocarbons and the long-range transport of ozone and PAN across Europe, Atmos. Environment, 25A, Derwent, R.G.; Hov, O. J. (1988): Geophys. Res., 93, Dunker, A.M. (1981): Atmos. Environ, 15, Dunker, A.M. (1984): J. Chem. Phys., 81, Easter R. C. (1993): Two modified versions of Bott s positive definite numerical advection scheme. Monthly Weather Review 1221, Elterman, L. (1968): Visible and IR attenuation for altitudes to 50 Km, Air Force Cambridge Res. Lab., Bedford, MA, AFCRL , 285. ENVIRON (1998): User s Guide Comprehensive Air Quality Model with Extensions (CAMx) Version ENVIRON Corporation, Novato, California, November. Falls, A.H.; McRae, G.J.; Seinfeld, J.H. (1979): Int. J. Chem. Kinet., CI, Filyushkin, V.V., S. Madronich, G.P. Brasseur, and I.V. Petropavlovskikh (1994): Fast twostream method for computing diurnal-mean actinic flux in vertically inhomogeneous atmospheres, J. Atmos. Sci., 51, Flatoy F. (1993): Balanced wind in advanced advection schemes when species with long lifetimes are transported. Atmospheric Environment 27, Freeman, D.; Egami, T.T.; Robinson, N.F.; Watson, J.G. (1986): J. Air Poll. Control Assoc., 36, Gao, D.; Stockwell, W.R. Milford, J.B. (1996): J Geophys. Res., 101, Gear C. W. (1971): Numerical Initial Value Problems in Ordinary Differential Equations, Prentice-Hall, Englewood Cliffs, NJ. Gelbard F. and Seinfeld J. H. (1980): Simulation of Multicomponent Aerosol Dynamics. Journal of Colloidal and Interfacial Science 78, Geron C. D., Guenther A. B. and Pierce T.E. (1994): An Improved Model For Estimating Emissions of Volatile Organic Compounds From Forests in the Eastern United States. Journal of Geophysical Research 99,

137 Gery, M.W.; Whitten, G.Z.; Killus, J.P.; Dodge, M.C. (1989): J. Geophys. Res., 94, Gery M. and R. Crouse, (1990): User s Guide for Executing OZIPR, Order No. 9D2196NASA, U.S. Environmental Protection Agency, Research Triangle Park, NC, Jang J. C. C., Jeffries H. E. and Tonnesen S. (1995b): Sensitivity of ozone to model grid resolution -- II: Detailed process analysis for ozone chemistry. Atmospheric Environment 29, Hales, J. M., M. W. Gery, and R. Crouse (1993): VOC Reactivity and its Application for Emission Regulation, Report submitted to the American Petroleum Institute. Hanna, S.R.; Chang, J.C.; Fernau, M.E. (1998): Atmos. Environ., 32(21), Hansen D. A., Dennis R., Ebel A., Hanna S. R., Kaye J. and Thullier R. (1994): The quest for an advanced regional air quality model. Environmental Science and Technology 28, Hestveldt E., Hov O. and Isaksen I.S.A. (1978): Quasi-steady-state approximations in air pollution modeling: A Comparison of Two Numerical Schemes For Oxidant Prediction. International Journal of Chemical Kinetics 10, Hov O., Zlatev Z., Berkowicz R., Eliason Z. and Prahm L.P. (1990): Comparison of Numerical Techniques for Use in Air Pollution Models With Non-Linear Chemical Reactions. Atmospheric Environment 23, Hough, A.M. and R.G. Derwent (1990): Changes in the global concentration of tropospheric ozone due to human activities, Nature, 344, 645. Houweling, S.H., F. Dentener, and J. Lelieveld (1998): The impact of nonmethane hydrocarbon compounds on tropospheric photochemistry, J. Geophys. Res., 103, 10,673. Hwang, J.T.; Dougherty, E.P.; Rabitz, S.; Rabitz, H. J. (1978): Chem. Phys., 69, Iman R.L.; Shortencarier, M.J. (1984): A FORTRAN 77 Program and Users Guide for the Generation of Latin Hypercube and Random Samples for Use with Computer Models; NUREG/CR-3624; Sandia National Laboratories: Albuquerque, NM, March. Intergovernmental Panel on Climate Change (1994): Radiative forcing of climate change and an evaluation of the IPCC IS92 emission scenarios, J.T. Houghton, L.G. Meira Filho, J. Bruce, H. Lee, B.A. Callander, E. Haites, N. Harris, and K. Maskell, eds.; Cambridge University Press. Isaksen, I.S.A. and O. Hov (1987): Calculation of trends in the tropospheric concentration of ozone, hydroxyl, carbon monoxide, methane and nitrogen oxides, Tellus, 39B,

138 Jacob D. J. (1998): Heterogeneous chemistry and tropospheric ozone. Submitted to Atmospheric Environment, special issue, 1998 NARSTO Assessment. Jacobson M. Z. and Turco R. P. (1994): SMVGEAR: A Sparse-Matrix, Vectorized GEAR code for atmospheric models. Atmospheric Environment 28, Jacobson M. Z. (1997): Development and application of a new air pollution modeling system --- Part II: Aerosol module structure and design. Atmospheric Environment 31, Jeffries, H.E. and Crouse, R. (1991):. Scientific and Technical Issues Related to the Application of Incremental Reactivity. Part H: Explaining Mechanism Differences, Report prepared for Western States Petroleum Association, Glendale, CA (October). Jeffries, H. E. and G. S. Tonnesen (1994): Comparison of Two Photochemical Reaction Mechanisms using a Mass Balance and Process Analysis, Atmospheric Environment, 28, Kahn, M.; Yang, YJ; and Russell, A.G. (1997): "Reactivities of Organic Solvents: Comparison between Urban and Regional Domains," Paper No. WA-97-!1540. Presented at the A&WMA Annual Meeting, Toronto, Canada. June. Air & Waste Management Association, Pittsburgh, PA. Kanakidou, M. and P.J. Crutzen (1993): Scale problems in global tropospheric chemistry modeling: comparison of results obtained with a three-dimensional model, adopting longitudinally uniform and varying emissions of NOx and NMHC, Chemosphere, 26, 787. Kerr, J. A. and J. G. Calvert (1985): Chemical Transformation Modules for Eulerian Acid Deposition Models. Volume 1: The Gas phase Chemistry, EPA-600/ , U.S. Environmental Protection Agency, Research Triangle Park, NC. King, M.D. and Harshvardhan (1986): Comparative accuracy of selected multiple scattering approximations, J. Atmos. Sci., 43, Kleffmann, J., K. H. Becker and P. Wiesen (1998): Heterogeneous NO2 conversion processes on acid surfaces: possible atmospheric implications, Atmos. Environment, 32, Kleinman L. I., P. Daum, J. H. Lee, Y.-N. Lee, L. Nunnermacker, S. Springston, J. Weinstein- Lloyd, L. Newman, and S. Sillman (1997): Dependence of ozone production on NO and hydrocarbons in the troposphere, Geophys. Res. Letters, 24, Kleinman L. I., (1994): Low- and High-NOx tropospheric photochemistry, J. Geophys. Res., 99, 16,831-16,838. Kumar N. and Russell A. G. (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. Journal of Geophysical Research 101,

139 Kuo, J.L. (1974): Further studies of the parameterization of the influence of cumulus convection on large-scale flow, J. Atmos. Sci., 31, Koda, M.; McRae, G.J.; Seinfeld, J.H. (1979): Int. J. Chem. Kinet., 11, Lacis, A.A. and J.E. Hansen (1974): A parameterization for the absorption of solar radiation in the earths atmosphere, J. Atmos. Sci., 31, Lelieveld, J. and P. J. Crutzen (1990): Influences of cloud photochemical processes on tropospheric ozone, Nature, 343, Lu, Ci-H, J. S. Chang (1998): On the indicator-based approach to asses ozone sensitvities and emissions features, J. Geophys. Res., 103, Lo Cha-Yang S. and H. E. Jeffries (1997): A quantitative technique for assessing pollutant source location and process composition in photochemical grid models, 90th AWMA Meeting, Toronto, Ontario, CA. Madronich, S. (1987): Photodissociation in the atmosphere, 1. Actinic flux and the effects of ground reflections and clouds, J.Geophys. Res., 92, Martinez, R.D., A.A. Buitrago, N.W. Howell, C.H. Hearn, and J.A. Joens (1992): The near U.V. absorption spectra of several aliphatic aldehydes and ketones at 300K, Atmos. Environ., 26A, McNair, L.A.; Harley, R.A.; Russell, A.G. (1996): Atmos. Environ., 30(24), McRae G. J., Goodin W. R. and Seinfeld J. H. (1982): Numerical solution of the atmospheric diffusion equation for chemically reacting flows, J. Computational Physics, 45, Milford, J., D. Gao, S. Sillman, P. Blossey, and A. G. Russell (1994): Total reactive nitrogen (NOy) as an indicator for the sensitivity of ozone to NOx and hydrocarbons, J. Geophys. Res., 99, Milford, J.B.; Gao, D. Russell, A.G.; McRae (1992): G.J. Environ. Sci. Technol., 26, Modica, L.G., D.R. Dulleba, R.A. Walters, and J.E. Langstaff (1991): Flexible Regional Emissions Data System (FREDS) Documentation for the 1985 NAPAP Emissions Inventory. EPA-600/ ORD, US EPA, May National Research Council (1991): Rethinking the ozone problem, National Academy Press, Washington, DC. 41

140 Noblet, G.; Harley, R.; Dhieux, J.; Milford, J. (1998): Effect of Wind Field Uncertainty on Predictions of a Lagrangian Photochemical Air Quality Model, 98-RA76B.02, Proceedings of the 91st Annual Meeting of the Air & Waste Management Association, San Diego, CA, June Odman M. T., Wilkinson J. G., McNair L. A., Russell A. G., Ingram C. L., and Houyoux M. R. (1996): Horizontal advection solver uncertainty in the Urban Airshed Model. Report to California Air Resources Board, Sacramento, California. Odman M. T., Kumar N. and Russell A. G. (1992): A comparison of fast chemical kinetic solvers. Atmospheric Environment 26, Odman, M. T. and A. G. Russel (1991): Multiscale Modeling of Pollutant Transport and Chemistry, J. Geophys. Res., 96, Olaguer, E.P. (1998): A global 3-D radiative-dynamical-chemical model for determining largescale impacts of atmospheric ozone precursors, in Proceedings of the Photochemical Reactivity Workshop, U.S. Environmental Protection Agency, May 12-14, Durham, NC. Oreskes N., Shrader-Frechette K. and Belitz K. (1994): Verification, validation, and confirmation of numerical models in the earth sciences. Science 263, Pandis, S.N.; Seinfeld, J.H. (1989): J. Geophys. Res., 94, Pierson W. R., Gertler A. W. and Bradow R. L. (1990): Comparison of the SCAQS tunnel study with other on-road vehicle emissions data. Journal of Air &Waste Management Association 40, Pleim J. E. and Chang J. S. (1992): A nonlocal closure model for vertical mixing in the convective boundary layer. Atmospheric Environment 26, Pierce, T., C. Geron, L. Bender, R. L. Dennis, G. S. Tonnesen, and A. Guenther (1998): Influence of isoprene emissions on regional ozone modeling, J. Geophys. Res., 103, 25, Pierce, T., B. Lamb, and A. Van Meter (1990): Development of a biogenic emissions inventory system for regional scale air pollution models, Proceedings of the 83rd A&WMA meeting, Pittsburgh, Pennsylvania, June Pudykiewicz, J. A., A. Kallaur, and P. K. Smolarkiewicz (1997): Semi-Lagrangian modeling of tropospheric ozone, Tellus, 49B, Radian Corporation (1993): GEMAP system documentation. Sacramento, CA. Draft, May. 42

141 Reynolds, S. And P. Roth (1997): Peer Review of the CAMx Ozone Source Apportionment Technology. Report from the EPA Source Attribution Workshop, July 16-18, 1997, RTP, NC. U.S. Environmental Protection Agency, Research Triangle Park, North Carolina. Roberts, J.M. and R.W. Fajer (1989): UV absorption cross sections of organic nitrates of potential atmospheric importance and estimation of atmospheric lifetimes, Environ. Sci. Technol., 23, Roelofs, G.-J., J. Lelieveld, and R. van Dorland (1997): A three-dimensional chemistry/general circulation model simulation of anthropogenically derived ozone in the troposphere and its radiative forcing, J. Geophys. Res., 102, 23,389. Rood R. B. (1987): Numerical Advection Algorithms and Their Role in Atmospheric Chemistry Transport Models, Review of Geophysics, 25, Roth P. M., Tesche T. W. and Reynolds S. D. (1998): A critical review of regulatory air quality modeling for tropospheric ozone. Submitted to Atmospheric Environment, special issue, 1998 NARSTO Assessment. Russell, A.G.; Dennis, R. (1998): NARSTO Critical Review of Photochemical Models and Modeling, submitted to Atmos. Environment. Russell G. L. and Lerner J. A. (1981): A new finite differencing scheme for the tracer transport equation. Journal of Applied Meteorology 20, Sandu A., Verwer J.G., Blom J. G., Spee E. J., Carmichael G. R. and Potra F. A. (1997) Benchmarking stiff ODE solvers for atmospheric chemistry problems 2. Rosenbrock solvers. Atmospheric Environment 31, Saylor R. D. and Ford G. D. (1995): On the comparison of numerical methods for the integration of kinetic equations in atmospheric chemistry and transport models. Atmospheric Environment 29, Sillman, S. (1998): NARSTO Review Article: The method of photochemical indicators as a basis for analyzing O3-NOx-ROG sensitivity, submitted to Atmos. Environment. Sillman, S., Dongyang He, M. R. Pippin, P. H. Daum, J. H. Lee, L. I. Kleinman, and J. Weinstein-Lloyd, (1997a): Model correlations for ozone, reactive nitrogen and peroxides for Nashville in comparison with measurements: Implications for O3-NOx-hydrocarbon chemistry, J. Geophys. Res. (submitted). Sillman, S., D. He, C. Cardelino and R. E. Imhoff (1997b): The use of photochemical indicators to evaluate ozone-nox-hydrocarbon sensitivity: Case studies from Atlanta, New York and Los Angeles, J. Air Waste Management Assoc. 43

142 Sillman, S. (1995): The use of NOy, H2O2, and HNO3 as indicators for ozone-noxhydrocarbon sensitivity in urban locations, J. Geophys. Res., 100, Sillman, S., F. J. Al-Wali, P. Marsik, P. Nowatski, P. J. Samson, M. O. Rodgers, L. J. Garland, J. E. Martinez, C. Stoneking, R. E. Imhoff, J-H. Lee, J. B Weinstein-Lloyd, L. Newman and V. Aneja (1995): Photochemistry of ozone formation in Atlanta, GA: models and measurements, Atmos. Environment, 29, Sillman S., Logan J. A. and Wofsy, S. C. (1990a): A regional-scale model for ozone in the United States with a sub-grid scale representation of urban and power plant plumes. Journal of Geophysical Research 95, Simpson et al. (1997): Photochemical oxidant modeling in Europe: multi-annulal modeling and source receptor relationships, Norwegian Meteorlogical Institute, EMEP MSC-W Report 3/97. Simpson D. (1995a): Biogenic emissions in Europe 2: Implications for ozone control strategies. Journal of Geophysical Research 100, 22,891-22,906. Simpson, D., (1995b): Hydrocarbon reactivity and ozone formation in Europe, J. Atmos. Chemisty, 20, Simpson D. (1992): Long period modelling of photochemical oxidants in Europe: Calculations for July Atmospheric Environment 26, Smolarkiewicz, P.K. (1983): A simple positive definite advection scheme with small implicit diffusion, Mon. Wea. Rev., 111, 479. Stevens, P. S., Mather, J. H., Brune, W. H., Eisele, F., Tanner, D., Jefferson, A., Cantrell, C., Shetter, R., Sewall, S., Fried, A., Henry, B., Williams, E., Bagman, K., Goldan, P., Kuster, W. (1997): HO2/OH and RO2/HO2 ratios during the Tropospheric OH Photochemistry Experiment: Measurement and theory, J. Geophys. Res., 102, Stockwell, W.R.; Middleton, P.; Chang, J.S.; Tang, X. (1990): J. Geophys. Res., 95, Stockwell, W.R., F. Kirchner, M. Kuhn, and S. Seefeld (1997): A new mechanism for regional atmospheric chemistry modeling. J. Geophys. Res., 102, Stockwell, W.R.; Yang, Y-J.; Milford, J.B. (1994): A Compilation of Estimated Uncertainty Factors for Rate Constants in Carter s Detailed Mechanism. Report to the Auto/Oil Air Quality Improvement Research Program, CRC, Atlanta, GA. Tie, X.-X., F.N. Alyea, D.M. Cunnold, and C.-Y. Jim Kao (1991): Atmospheric methane: a global three-dimensional model study, J. Geophys. Res., 96, 17,

143 Tesche T. W., Lurmann F. L., Roth P. M., Georgopoulos P., Seinfeld J. H. and Cass G. (1990): Improvement of Procedures for Evaluating Photochemical Models. Report to California Air Resources Board for Contract No. A Alpine Geophysics, LLC, Crested Butte, Colorado. Thompson, A.M.; Stewart, R.W. (1991): J. Geophys. Res., 96, Tonnesen, G. S. (1998): Effects of Uncertainty in the Reaction of the Hydroxyl Radical with Nitrogen Dioxide on Model Simulated Ozone Control Strategies, Atmos. Environment (in press). Tonnesen, G. S. and R. L. Dennis (1998a): Analysis of radical propagation efficiency to assess ozone sensitivity to hydrocarbons and NOx. Part 1: Local indicators of odd oxygen production sensitivity, (submitted to J. Geophys. Res.). Tonnesen, G. S., and R. L. Dennis, (1998b): Analysis of radical propagation efficiency to assess ozone sensitivity to hydrocarbons and NOx. Part 2: Long-lived species as indicators of ozone concentration sensitivity, (submitted to J. Geophys. Res.). Tonnesen, G. S., R. L. Dennis, G. Gipson (1998): Estimating the Area of Influence of Ozone Produced by Local Precursors Emissions, In: Preprints of the 10th Joint Conference on the Applications of Air Pollution Meteorology with the Air & Waste Management Association, January 11-16, 1998, Phoenix, Arizona. American Meteorological Society, Boston, Massachusetts, paper no. 9A.15, pp Tonnesen, S., and H. E. Jeffries (1994): Inhibition of Odd Oxygen Production in the Carbon Bond Four and Generic Reaction Set Mechanisms, Atmos. Environment, 28, U. S. Environmental Protection Agency, Office of Air and Radiation, Office of Air Quality Planning and Standards (1993): User s guide for the Urban Airshed Model. Volume IV: User s manual for the emissions preprocessor system. USEPA Report No. EPA-450/ D (June 1990 edition revised March ) U. S. EPA, Office of Air Quality Planning and Standards, Research Triangle Park, North Carolina. Wang, Z. (1998): Computing Volatile Organic Compound Reactivities with a 3-D AQM, in Proceedings of the Photochemical Reactivity Workshop, U.S. Environmental Protection Agency, May 12-14, Durham, NC. Wang, Y., D.J. Jacob, and J.A. Logan (1998): Global simulation of tropospheric O3-NOxhydrocarbon chemistry. 1. Model formulation, J. Geophys. Res., 103, 10,713. Whitby E. R., McMurray P. H., Shankar U. and Binkowski F.S. (1991): Model Aerosol Dynamics Modeling.U. S. EPA Report No. 600/3-91/020. U.S. Environmental Protection Agency, Research Triangle Park, North Carolina. 45

144 Wilkinson J. G., Loomis C. F., McNally D. E., Emigh R. A. and Tesche T. W. (1994): Technical Formulation Document: SARMAP/LMOS Emissions Modeling System (EMS-95). Alpine Physics Document No. 90/TS26&90/TS27. Alpine Geophysics, Pittsburgh, Pennsylvania. Yang, et al., (1999) to be submitted. Yang, Y-J.; Stockwell, W.R.; Milford, J.B. (1995): Environ. Sci. Technol, 29, Yang, Y-J. and Milford, J.B. (1996a): Environ. Sci. Technol., 30, Yang, Y-J.; Stockwell, W.R.; Milford, J.B. (1996b): Environ. Sci. Technol., 30, Yang, Y-J.; Wilkinson, J.G. ; Russell, A.G. (1997): Environ. Sci. Technol., 31, Yarwood, G., R.E. Morris, M.A. Yocke, H. Hogo and T. Chico (1996): Development of a methodology for source apportionment of ozone concentration estimates from a photochemical grid model, 89th AWMA Annual Meeting, Nashville, TN, June Young T. R. and Boris J. P. (1977): A numerical technique for solving ordinary differential equations associated with the chemical kinetics of reactive flow problems. Journal of Physical Chemistry 81, Zannetti, P. (1990): Air Pollution Modeling, Van Nostrand Reinhold, New York, 444 pp. 46

145 TABLES Table 1U. Recent uncertainty analysis studies of air quality models and of reactivity quantification. Air Quality Model Chemical Mechanis Uncertain Output Parameters Evaluated Analysis Approach Ref. UAM threedimensional CIT trajectory version CIT threedimensional CIT threedimensional Box model m CB-IV (Gery et al., 1989) SAPRC93 (Carter, 1995) SAPRC90 (Carter, 1990) SAPRC90 RADM2 (Stockwell et al., 1990) Evaluated O 3 O 3, HCHO, HNO 3, PAN, NO y, and motor vehicle emissions reduction scenarios O 3 and VOC reactivity quantificati on O 3 and VOC reactivity quantificati on O 3, HCHO, H 2 O 2, and PAN; ozone control strategies 23 variables related to emissions, boundary conditions, and meteorological conditions; 86 chemical rate constant variables. Total: 109 parameters. Trajectory path, reflecting uncertainties in wind speed and direction; 29 chemical rate constant variables, 3 deposition affinities, 11 mixing height periods, the atmospheric stability class, and 6 classes of emissions. Total: 51 parameters. 12 reaction rate constants, emissions, mixing heights. 47 Uncertainty estimates through survey of 10 modeling experts. Uncertainty analysis of ozone prediction performed using 50 Monte Carlo simulations. Uncertainty estimates though published review (chemistry), sensitivity studies, data withholding analysis, and subjective estimates. Uncertainty analysis of ozone and other secondary pollutant predictions and of ozone control strategy evaluation performed using sets of 200 Monte Carlo simulations. Studied uncertainties in VOC reactivity quantification, ozone predictions and control strategy effectiveness extending the work of Bergin et al., (1998, 1999). 9 reaction rate constants. Uncertainty estimates extending previous box model study (Yang et al., 1994). Sensitivity analysis of chemical mechanism rate constant uncertainties on reactivity quantification for 26 organic compounds and carbon monoxide. Reaction rate constants and product yields. Uncertainty estimates from previous published study. Global Monte Carlo calculations of secondary pollutant formation and ozone Hanna et al., 1998 Bergin et al., 1998a Yang et al., 1999 Bergin et al., 1998c 48, 14 Gao et al, 1996; Yang et al, 1996b

146 Box model SAPRC90 O 3 and VOC incremental reactivity quantificati on Box model SAPRC90 Reactivity Adjustment Factors from alternative fuels CIT Threedimensional SAPRC90 O 3 and Reactivity Adjustment Factors from 5 alternative fuels control strategies. 73 rate parameters Uncertainties estimated though expert and published review. Initial first order sensitivity analysis of 201 compounds using the Direct Decoupled Method (DDM; ref). Uncertainties in incremental reactivities calculated using sets of 100 Monte Carlo simulations. 34 organic compound Uncertainties from incremental reactivities previous published studies. and emissions Analysis using 300 Monte composition of 8 fuels Carlo simulations. 11 reaction rate constants and one product yield parameter. Uncertainties from Yang et al., 1996a. Sensitivity analysis of chemical mechanism rate constant and product yield uncertainties on quantification of Reactivity Adjustment Factors for CNG, LPG, Ph2 gasoline, M85, and E85. Yang et al., 1995 Yang et al., 1996a Bergin et al.,

147 Table 1. Current Photochemical Air Quality Models and Attributes. The list is meant to be instructive as to the variety and history of photochemical models, and is not exhaustive. a. Urban Scale Model UAM-IV Full Name/ Reference Urban Airshed Model, version IV (Reynolds et al., 1973) Model Type Eulerian, multilayer, non-nested grid Chemical Mechanism (often used: most can use a variety of mechanisms) CB-IV Grid size, Features, Comments Variable horizontal grid size (typically about 5 km). Typical Applications (Location/ Pollutant/ research and/or policy focus) North America., Europe/ U.S. Regulatory Model CIT CALGRID EKMA California/Ca rnegie Institute of Technology Model (Harley et al., 1992) California Air Resources Board Grid Model (Yamartino et al., 1992) Empirical Kinetics Modeling Approach (NRC, 1991) Eulerian, multilayer, non-nested grid (Trajectory form available.) Eulerian, multilayer, non-nested grid Box Model (including trajectory form) SAPRC90/93 (readily changed) CB4, SAPRC90/93 (readily changed) Variable (CB4, SAPRC, RADM, etc.) variable grid (typically about 4-5 km.), cloud processes, dry dep, aerosol dynamics, sensitivity anal. variable grid (typically about 4-5 km.), cloud processes, dry dep, Little physical detail. Used for chemical kinetics investigatio n U.S, Europe, Asia, Australia/Ozon e, PM & deposition & Ozone/ Policy & research U.S, /Ozone, PM & deposition & Ozone/ Policy & research Historical model used early in U.S. for policy and research. Used for studies focussing on chemistry Table is taken from Russell and Dennis (1998). 49

148 b. Regional Scale Model Full Name/ Reference STEM/ STEM II RADM ROM EMEP Sulfur Model EMEP Ozone Model Sulfur Transport and Emissions Model (Carmichael et al., 1991) Regional Acid Deposition Model (Chang et al., 1987) Regional Oxidant Model (Lamb, 1983) European Monitoring and Evaluation Programme Sulfur Model (Eliassen et al., 1982) European Monitoring and Evaluation Programme Oxidant Model Type Eulerian, multilayer, non-nested grid Eulerian, multilayer, non-nested grid Eulerian, 3 layer, nonnested grid Lagrangian, 1 Layer Lagrangian, 1 Layer Chemical Mechanism (often used: most can use a variety of mechanisms) Atkinson et al. (1992) RADM2 (Stockwell et al., 1990) CB-IV (Gery et al., 1989) EMEP EMEP Grid size, Features, Comments Variable horizontal grid size (typically about 50 km)., cloud process, wet and dry deposition variable grid (typically about 80 km.), cloud processes, wet and dry dep. Typical horiz. grid about 18 km. 150 km resolution. Typical Applications (Location/ Pollutant/ research and/or policy focus) U.S. and Asia/ Acid deposition/ Research U.S/Acid deposition & Ozone/ Policy & research U.S./ Ozone/ Policy & research Europe/Ozo ne/ Policy & research Europe/Ozo ne/ Policy & research Table is taken from Russell and Dennis (1998). 50

149 EURAD LOTOS Harvard Model Model (Simpson, 1992) European Air Dispersion Model (Hass, 1991) Long Term Ozone Simulation Model (Builtjes et al., 1988) (Sillman et al., 1990a) Eulerian, multilayer, nesting possible Eulerian, 3 layer, nonnested grid Eulerian, Multilayer RADM2 CBM-IV Harvard Similar to RADM/SAQ M Grid 0.5 o Lat x 1.0 o Lon. Europe/Acid deposition and ozone/ Research Europe/Ozo ne/ Research U.S./Ozone/ Research 51

150 Table 1 (cont.). Current Photochemical Air Quality Models and Attributes NOAA National Oceanic and Atmospheric Administrati on (McKeen et al., 1991a) Eulerian, Multilayer NOAA REM-3 ADOM Regional Eulerian Model with 3 chemistry schemes (Stern, 1994) Acid Deposition and Oxidant Model (Venkatram et al., 1988) Eulerian, 3 layers Eulerian, Multilayer CB-IV, SAPRC-90, Harwell Atkinson et al. (1992) Grid 0.25 o Lat x 0.5 o Lon. Variable grid size U.S./Ozone/ Research Europe/Ozon e/ Policy & Research North America Ozone/Policy c. Multiscale/Nested Models Model SAQM MAQSIP Full Name/ Reference SARMAP Air Quality Model (Chang et al., 1997) Multiscale Air Quality Simulation Program (Odman and Ingram, Model Type Eulerian, multilayer, nested grid Eulerian, multilayer, nested grid Chemical Mechanism (often used: most can use a variety of mechanisms) CBM-IV CBM-IV Grid size, Features, Comments Grid size from km. Grid size from km. Typical Applications (Location/ Pollutant/ research and/or policy focus) U.S./Ozone/ Policy (New model) Table is taken from Russell and Dennis (1998). 52

151 EURAD UAM-V URM 1996) European Air Dispersion Model (Hass, 1991) Urban Airshed Model- Variable (Morris, et al., 1991) Urban-to- Regional Multiscale Model (Kumar et al., 1994) Eulerian, multilayer, nesting possible Eulerian, multilayer, nested grid Eulerian, multilayer, multiscale RADM2 CB-IV SAPRC (e.g., Carter, 1990) Similar to RADM/SAQ M Grid sizes range from 4-50+km., Plume-in-grid Grid sizes from 4-200km, Plume-in- Grid, aerosol dynamics Europe/Acid deposition and ozone/ Research U.S./Ozone/ Policy U.S./Ozone/ Research and Policy d. Photochemical models with aerosol dynamics routines. * Model Full Name/ Reference Aerosol Treatment UAM-AERO (See UAM description above) Urban Airshed Model, Aerosol (Lurmann et al., 1997) Sectional approach CIT-AERO (See CIT description above) RPM (Based on RADM, described above) SAQM-AERO (Based on SAQM, described above) URM (See URM description above) CIT Aerosol Regional Particulate Model (Binkowski and Shankar, 1995) SAQM, Aerosol Urban-to-regional Multiscale (Kumar et al., 1994) Sectional approach Modal approach Sectional approach Sectional approach Table provided by Robin Dennis, personal communication (1998). 53

152 FIGURES Figure 1. Non-Linearity and error in [O3] and d[o3]/denox with size of emissions increment d[o3] (ppb) % -3.00% -2.00% -1.00% 0.00% 1.00% 2.00% 3.00% 4.00% %denox ErrTol= ErrTol= errtol = d [O3]/dENOx (ppb/ppb) % -3.00% -2.00% -1.00% 0.00% 1.00% 2.00% 3.00% 4.00% %denox ErrTol= ErrTol=

153 Figure 2. Non-linearity in HCHO incremtal reactivity as a function of emissions increment Incremental Reactivity % % % % % 0.00% 20.00% 40.00% 60.00% 80.00% % % change in HCHO emissions Figure 3. Sensitivities of O3 Production to NO and VOC mixing ratios. Results expressed in ppbv change in O 3 /percent change in initial concentration. 55

154 REACTIVITY ASSESSMENTS 5/5/99 M. Bergin 1, W.P.L. Carter 2, J. Milford 1, Philip J. Ostrowski 3 and A.G. Russell 1 1 Georgia Institute of Technology, Atlanta, Ga 2 University of California, Riverside 3 Occidental Chemical Co. CONTENTS QUANTIFICATION OF VOC REACTIVITY...1 Modeling-based Assessments of VOC Reactivity...4 Trajectory Model Reactivity Assessments...5 Eulerian Model Reactivity Assessment...7 VARIABILITY IN REACTIVITY SCALES...10 CHEMICAL MECHANISM UNCERTAINTY...14 REFERENCES...17 TABLES...22 FIGURES...29 Note: Much of the text on Quantification of VOC Reactivity and Variability in Reactivity Scales is from Bergin et al. (1998). Authors names are in alphabetical order. QUANTIFICATION OF VOC REACTIVITY A number of possible methods to quantify the impact of a VOC on ozone formation have been examined. Early reactivity experiments were based on amounts of ozone formed when the VOC is irradiated in the presence of NO x in environmental chambers (e.g., Wilson and Doyle, 1970; Altshuller and Bufalini, 1982; Laity et al, 1973). However, individual VOCs are not emitted in the absence of other reactive organics, so such experiments cannot be expected to represent atmospheric conditions. Furthermore, there are chamber wall and background effects which affect the results of such experiments, particularly if the compound reacts slowly or has radical sinks in its mechanism (Carter and Lurmann, 1990, 1991; Carter et al, 1982; Bufalini et al, 1977; Joshi et al, 1973; Jeffries and Sexton, 1993). An alternative measure that has been proposed is comparison of the OH radical rate constants between VOCs (e.g., Darnall et al, 1976; CARB, 1989; Chameides et al, 1992). The EPA has for the most part has used the OH rate constant (with ethane as the boundary) for exempting compounds from VOC regulations 1

155 (Dimitriades, 1996). Although not strictly a measure of ozone formation, for most compounds reaction with OH is the main process that initiates the VOC's ozone-forming reactions. This approach has the significant advantage that the OH rate constants are known or can be estimated for essentially all VOCs of relevance to most regulatory applications (Atkinson, 1987, 1989, 1990, 1994), and the OH rate constants are properties only of the VOC, and not the environment where the VOC is emitted (other than small temperature dependencies.) However, a significant number of compounds also react nonnegligibly with ozone, NO 3 radicals or by direct photolysis, and also this method does not account for the significant differences in the subsequent reaction pathways of the initial products, which can affect how much ozone is formed after the VOC-OH reaction OH (Carter and Atkinson, 1989; Carter, 1994). In particular, model calculations have shown that incremental reactivities of VOCs depend not only on how fast the VOC reacts, but also on the tendency of the VOC to enhance or inhibit radical levels, the tendency of the VOC to remove NO x from the system, and the reactivity of the VOC's major products (Atkinson, 1990; Carter and Atkinson, 1989; Bowman and Seinfeld, 1994a, 1995). For example, aromatics, which have strong NO x sinks and radical sources in their mechanisms, are predicted to have relatively high reactivities under low VOC/NO x conditions, but are predicted to have negative reactivities when the VOC/NO x ratio was sufficiently high. For this reason, the OH radical rate constant has been shown to correlate poorly with other measures of ozone formation potential, particularly for the more rapidly reacting VOCs (e.g., Dodge, 1994, Carter, 1991). Because of the limitations of the OH radical rate constant scale, Weir et al. (1988), and Carter (1991) argued that a scale based on incremental reactivities would provide a more comprehensive measure of the effect of a VOC on actual ozone formation. This is defined as the change in ozone caused by a change in the emissions of a VOC in a specific air pollution episode. To remove the dependence on the amount of VOC added, incremental reactivity is defined by Equation 1 as the limit as the amount of VOC added approaches zero, i.e., as the derivative of ozone with respect to VOC (as shown on the lower plots of Figure 1): IR i [ O ] [ VOC ] = 3 i (1) Here, IR i is the incremental reactivity and the subscript i denotes the VOC being examined. This reactivity definition takes into account the effects of all aspects of the organic's reaction mechanism and the effects of the environment where the VOC is emitted. However, model calculations (Dodge, 1977; Carter, 1991, 1994; Chang and Rudy, 1990) and environmental chamber experiments (Carter et al, 1995a) have shown that changes in environmental conditions can significantly affect incremental reactivities, both in a relative and in an absolute sense. Therefore, the incremental reactivity is a function of the episode as well as the VOC. This presents obvious problems in developing reactivity scales for use in VOC control regulations that will be applicable under all conditions. Methods for dealing with this episode dependence are discussed in the section on variability below. 2

156 The incremental reactivity of a VOC under true ambient conditions cannot be measured directly other than by changing emissions and then observing the resulting changes of air quality for enough years to factor out effects of meteorological variability but can be estimated either by computer model calculations or by suitably designed environmental chamber experiments. Both types of estimation approaches have their limitations. In the case of model calculations, uncertainties and approximations in the model for airshed conditions, in the model formulation, and in the chemical mechanism cause uncertainties in the predicted ozone impacts, as discussed further below. In the case of experiments, it is difficult for the conditions of the experiment to simulate ambient conditions, so the results do not have general applicability. For these reasons, modeling and experimental measurements are used in conjunction for examining reactivity. In an early model simulation, Dodge (1984) showed that, when adding a given amount of a VOC to the other VOC inputs in EKMA model simulations, the calculated change in ozone varied widely among different VOCs at low VOC/NO x ratios, but were lower and less variable under high VOC/NO x conditions. Carter and Atkinson (1989) showed that while the VOC/NO x ratio was probably the most important single environmental factor affecting reactivity, other factors are important as well. Simulations of environmental chamber experiments resulted in different incremental reactivities (both absolute and relative) than simulations of atmospheric conditions, indicating that incremental reactivities measured in chamber experiments should not be used to assess atmospheric reactivities without the benefit of model calculations to account for the differences between chamber and airshed conditions. In addition, Carter and Atkinson (1989) showed that the number of days in the pollution episode and the nature of the other VOCs present also had a non-negligible effect on VOC reactivities. Carter (1994) showed that there was still some variability in relative reactivities among different one-day airshed model scenarios even after NO x inputs in the scenarios were adjusted to yield consistent conditions of NO x availability. Jeffries and Crouse (1991) and Bowman and Seinfeld (1994a,b, 1995) looked at the factors affecting reactivity from the perspective of the chemical reactions actually responsible for ozone formation. The result was that the relative contribution of VOCs to the reactions that are directly responsible for ozone could be quite different than the relative incremental reactivities of those VOCs. This is because many VOCs have high (or negative) incremental reactivities not because of the ozone formed by their own reactions, but because their reactions affect how much ozone is formed from other VOCs. For example, if the reactions of a VOC significantly affect radical levels they will affect how much O 3 is formed from the reactions of other VOCs, and for many VOCs this indirect effect on reactivity makes a larger contribution to its incremental reactivity than the ozone formed by the VOC's direct reactions (Bowman and Seinfeld, 1994a,b, 1995). This result has also been shown from an analysis of the results of incremental reactivity experiments carried out under maximum reactivity conditions (Carter et al, 1993, 1995b). A general scale that ranks the reactivities of VOCs would clearly aid the development of regulatory applications that take differences in VOC reactivity into account. However, because incremental reactivities depend on environmental conditions, no incremental reactivity scale will correctly predict relative ozone impacts under all conditions (even if there were no uncertainties in the models, the chemical mechanism, and the airshed conditions.) This can be partially 3

157 accounted for through the use of a relative, rather than absolute, comparison of reactivities. In other words, we do not compare the absolute amount of ozone formed per amount of VOC added, but the amount of ozone formed relative to other VOCs. This concept is applied to sources as well as compounds. For example, if one is comparing the reactivity of emissions from a gasoline-fueled vehicle to that from a compressed natural gas (CNG) vehicle, what is most important is not that the gasoline-fueled exhaust has a reactivity of 1.00 gram of ozone per gram of exhaust VOC, and CNG exhaust has a reactivity of 0.20 grams of ozone per gram exhaust VOC. These quantities are dependent on location and time. What is of greater interest is that the CNG exhaust is 5 times less reactive, so the CNG vehicle can emit about five times as much VOC in any area and still have a similar impact on ozone levels. Defining reactivity in reference to other reactivity values rather than to absolute ozone formation allows reactivity values to be more readily evaluated and compared. To calculate relative compound reactivities, we quantify the reactivity of individual VOCs as compared to a reference compound or, better, a VOC mixture (i.e., the reactivities are normalized). When these normalized compound reactivities are quantified in a relative sense, the set of reactivities is referred to as a relative reactivity scale. While the use of a reactivity scale reduces the effect of reactivity variability, it cannot completely account for environmental effects. Nevertheless, the only practical alternative to using a general reactivity scale is regulating all compounds as if they were either reactive or unreactive, i.e., using an implicit reactivity scale where all compounds have reactivities of either 0 or 1. This method has obvious shortcomings. For these reasons, a number of reactivity scales have been developed, and are summarized in Table 1. Modeling-based Assessments of VOC Reactivity Two methods are traditionally used to assess VOC reactivity: experiments using smog chambers and computationally using air quality models. Both have limitations. Smog chambers do not realistically represent the physics of pollutant transport and impact of fresh emissions, and are generally carried out at higher pollutant levels than generally occur in the atmosphere. Therefore, the conditions inside a smog chamber do not reflect those conditions found in the ambient air. Given the sensitivity of a compounds impact (and hence reactivity) to the environment, this limits the applicability of smog chamber experiments for reactivity quantification. In addition, chamber wall effects can also impair such experiments, particularly if the compound reacts slowly or has radical sinks in its mechanism (e.g., Carter and Lurmann, 1991). Models also have limitations. All models suffer from a limited knowledge in the chemistry of specific compounds, and this directly impacts one s ability to quantify the reactivity of that compound. For many compounds, the chemical mechanism is not well known. A second limitation is that other uncertainties lead to uncertainties in how a model treats the dynamics of various compounds. However, the models have been developed to accurately represent the various physical and chemical processes that impact the dynamics of trace gases in the atmosphere, so they provide a method to quantify reactivity in the atmosphere. Computational modeling has been the approach taken in most recent reactivity quantification studies (e.g., Carter and Atkinson, 1989; Carter, 1994; Derwent and Jenkins, 1991; Bowman and Seinfeld, 1995; McNair et al., 1992; Yang et al., 1995; Bergin et al., 1995). 4

158 The various types of model s used for reactivity assessments include box, trajectory and three-dimensional models (also called airshed models). The differences in these models are described in the section on modeling. The three-dimensional models have the fewest limitations in their formulation, and can also provide spatial information, e.g., how compound reactivities change over an airshed. Thus, they are the most powerful approach. On the other hand, they are the most computationally intensive and until recently could not be used with the more detailed chemical mechanisms which are required for reactivity assessment. In addition, the input data usually have major uncertainties, and if the input data are incorrect the model predictions may be no more accurate than those using simpler models. The simpler models (e.g., box and trajectory models) permit a wider variety of scenarios and more cases to be examined, in greater chemical detail and at lower cost. Simpler models were used for the first reactivity assessments, and the more recent assessments were conducted using multidimensional models applied at urban and regional scales. Both types of models, because of their individual strengths, continue to have a role in reactivity quantification. A list of model-based reactivity assessments is given in Table 1. Trajectory Model Reactivity Assessments The first set of reactivity assessments were conducted using single cell trajectory models so that a detailed chemical mechanism could be used and the model could be applied to a number of areas. Using this approach, Carter (1994) developed a total of 18 reactivity scales, all calculated using the SAPRC-90 chemical mechanism. Subsequent versions of these scales have been calculated by Carter using the updated (SAPRC-98) version of the mechanism (Carter, 1998), but the scenarios and reactivity assessment methods were not changed. The Carter scales represent the average results from 39 modeled trajectories, each representing a single day episode in an urban area with varying, though low, VOC-to-NO x ratios. Averaging reactivities across these trajectories accounts for some of the variability caused by environmental conditions, as discussed further below. The 18 reactivity scales were derived using nine different approaches for dealing with the dependence of reactivity on environmental conditions and on two methods for quantifying ozone impacts. A summary of these scales is presented in Table 2. Of those 18 scales, the Maximum Ozone Incremental Reactivity (MOIR) scale and the Maximum Incremental Reactivity (MIR) scale were found to be reasonably representative of the full set, and are discussed in more detail below. The MIR scale reflects primarily the effect of the VOC on ozone formation rates. The MOIR, Equal Benefit Incremental Reactivity (EBIR), and the base-case average ratio ozone yield scales are more sensitive to the effect of the VOC on ultimate O 3 yields in NO x -limited conditions. Scales based on integrated O 3 are sensitive to both factors, but tend to be more similar to MIR than MOIR (see also the discussion in following sections.) Scales sensitive to effects of VOCs on ozone formation rates generally give higher relative reactivities for aromatics, and lower relative reactivities for alkanes, than those based on ultimate O 3 yields in NO x -limited conditions. More recently, Carter (unpublished results) calculated reactivity scales based on the effect of the VOC on maximum 8-hour average ozone levels, which is more representative of 5

159 the new Federal ozone standard. The reactivities were similar to those calculated using the integrated ozone scale. Two of the above scales that have been most seriously considered for regulatory use are the MOIR scale and the MIR scale. The MOIR is based on incremental reactivities for NOx conditions which are most favorable to ozone formation, as indicated by the "MOR" point on the bottom-left plot in Figure 1. The MIR is based on the incremental reactivities of VOCs under relatively high NO x conditions where the VOCs have their highest incremental reactivity, as is also shown on the bottom-left plot of Figure 1. Carter and co-workers (Weir et al, 1988; Carter, 1994) proposed using the MIR scale for regulatory applications because the MIR scale reflects reactivities under environmental conditions which are most sensitive to effects of VOC controls. The MIR scale may be less accurate than others in predicting O 3 effects under lower NO x conditions; however, because of the lower sensitivity of O 3 under those conditions, the practical impact of those inaccuracies is less than would be the case for the conditions where the scale is designed to apply. The MIR scale was also found to correlate well to scales based on integrated O 3 yields, even in lower NO x scenarios. It can also be argued that this scale is appropriate when used in conjunction with a NO x control program, which provides the most effective ozone control under low NO x conditions. Nevertheless, the MOIR scale is attractive because it is more representative of the "worst case" ozone formation conditions in various airsheds, and also because it tends to be more conservative in predicting substitution benefits for most alternative fuels. The MIR scale tends to predict larger reactivity benefits for slowly reacting compounds than may be appropriate, because the higher NO x levels of MIR scenarios cause suppressed radical levels, which decrease the amount that slower reacting compounds react in the scenarios. Ultimately, CARB concluded that the MIR was a superior method to the OH scale for assessing reactivity, and used the scale as a basis for deriving Reactivity Adjustment Factors (RAFs) in California's LEV/CF regulations (CARB, 1991). RAFs are discussed further below. The MIR scale is now also widely used as a means for comparing reactivities of vehicle emissions during various driving cycles as well as with the use of alternative fuels (e.g., AQIRP, 1993). An alternative approach that may have the best features of both the MIR and MOIR would be to use a scale based on integrated ozone or maximum 8-hour average ozone under base-case or maximum ozone conditions. This has the advantage of the MIR scale in that it performs well in predicting reactivity effects under high NOx conditions that are most sensitive to VOCs (because it correlates reasonably well to MIR for most VOCs), while also being based on conditions of scenarios that are more representative of worst case O 3 pollution episodes. Furthermore, in the context of Eulerian model simulations where ozone impacts vary with both time and space, integrated ozone throughout the full air basin and time period of the episode is arguably a more robust measure of the exposure of the environment to ozone than the peak ozone concentration, which might be highly localized in time and place. Comparisons of Eulerian model predictions with the MIR and MOIR scales are discussed below. An alternative series of reactivity scales derived using a trajectory model are the Photochemical Ozone Creation Potential (POCP) scales, also shown on Table 1, which were calculated by Derwent and other researchers in Europe (e.g., Derwent and Jenkin, 1991; 6

160 Andersson-Skold et al, 1992) using updated versions of the Derwent and Hov (1979) chemical mechanism (see Atmospheric Chemistry section) and a two-layer Lagrangian model representing various multi-day trajectories across Europe. The reactivities are calculated from the change in mid-afternoon ozone for each day in the trajectory resulting from removing the test VOC from the emissions, divided by the integrated emissions of the test VOC up to the time of the ozone observation. Different POCP scales were calculated using different trajectories, and unlike the Carter approach no scales were derived to represent multiple conditions. Most of the POCP scenarios probably represent low NO x conditions. A comparison of MIR, MOIR, and POCP reactivities for selected VOCs is shown on Figure 2. The MIR and MOIR scales give very similar relative reactivities for most compounds, and are consistent in predictions of which compounds are highly reactive and which are not. However, for reasons indicated above, the MOIR scale gives lower relative reactivities for aromatics, and also predicts lower relative reactivities for radical initiators such as formaldehyde, which have larger effects on rates of ozone formation than on ultimate ozone yields. The MIR, MOIR, and POCP relative reactivities generally predict similar orderings of reactivities (relative reactivities), but some significant differences are observed. The largest differences, particularly for the alkanes and methyl ethyl ketone (MEK) are probably due primarily to differences in the chemical mechanisms employed, rather than the types of scenarios employed. The mechanisms used to develop the POCP reactivities are chemically detailed and intended to be explicit, but unlike the SAPRC and Carbon Bond mechanisms have not been evaluated against chamber data, and may not adequately represent the large NO x sink processes in the aromatic photooxidations that give them low or negative reactivities under low NO x conditions (see Atmospheric Chemistry section). The relatively low reactivity predictions for the higher alkanes by the SAPRC mechanisms have been verified by environmental chamber experiments (Carter and Lurmann, 1991; Carter et al, 1993, 1995b, 1997; Carter, 1995). Effects of differences and uncertainties in chemical mechanisms on reactivity scales are discussed in more detail in a later section. Jiang et al. (1996) also used a trajectory model to evaluate reactivity. This study used the SAPRC-90 chemical mechanism in the Ozone Isopleth Plotting Research version (OZIPR) trajectory model (Gery and Crouse, 1989) to predict the reactivity of 17 VOCs and methane in the Lower Fraser Valley of Canada. This study designated nine VOCs as significant contributors to the ozone concentrations, seven of which represent lumped compound groups. The greatest contributor to ozone formation was found to be ARO2, a lumped model species used to represent the xylenes and other fast reacting aromatics. Eulerian Model Reactivity Assessment A serious concern about the regulatory application of scales such as MIR and MOIR is that they are all based on the single cell (Lagrangian) model simulations of single-day pollution episodes. MIRs have been developed based on 10-hour simulations, whereas some organic compounds may remain in an urban airshed for 2 to 3 days, and even longer when considering regional ozone. Further, trajectory models lack the physical detail (e.g., wind shear, venting to/from the free troposphere), the spatial and temporal detail of emissions and resulting 7

161 pollutants, and the multi-day pollution effects that can be represented in Eulerian models. For that reason, it is important that the scales derived using trajectory models be evaluated using more detailed models with the same chemical mechanism. Initial studies applying threedimensional air quality models to assess reactivity have been carried out by Russell and coworkers (e.g. McNair et al, 1992; Bergin et al,1996,1998; Kahn et al, 1998a,b; Yang et al., 1998). Those studies primarily employed the Carnegie/California Institute of Technology (CIT) Model (Mc. Rae et al, 1982; Harley et al, 1993) applied to a 3-day air pollution episode in the Los Angeles air basin (Harley et al., 1994), or the Urban-to-Regional Multiscale (URM) model (Odman et al., 1994). In addition to Los Angeles-based studies, Kahn et al. (1998a) also applied the CIT model to the Swiss Plateau and Mexico City to study the use of reactivity over domains with greatly different levels of VOC and NO x, and Kahn et al. (1998b) studied reactivity variations over a regional domain, looking at the U.S.-Mexico border region. Two of these studies (Kahn et al. 1998b and Yang et al., 1998) incorporated the Direct Decoupled Method (DDM) for computational efficiency. A challenge in comparing VOC reactivity using results between box and grid modeling studies is the difference between quantification measures, or metrics, that can be defined from each analysis method. Differences in the spatial and temporal representation of emissions can also make the comparison of results difficult. In the Eulerian reactivity studies, incremental emissions of the test compound were modeled by increasing the test VOC proportionally to the spatial and temporal distribution of the base organic species emissions. The rates of all organic species emissions in each modeling cell for each hour were used to determine the rate of the test species emission in that cell. This is represented mathematically by Equation 2, where, at time t in model cell x, y, z, the perturbed emission (E p ) of test species i is calculated as the base emission of that species (E i b ) plus a fraction, (α), of the sum of the total base level emissions of reactive organic gases. Index j refers to each represented explicit or lumped emitted VOC. This modeling method accounts for the effect of emissions variation, transport, and multi-day reactions. E p i(x,y,z,t) = E b i(x,y,z,t) + ασ j E b j(x,y,z,t) (2) In addition to different representation of emissions by trajectory and three-dimensional models, results from three-dimensional modeling can be described in a number of ways. Three of the most useful metrics are the difference between peak ozone concentrations predicted using the base and perturbed inventories, and population- and exposure-weighted exposure to ozone levels exceeding a threshold value. Further description of these metrics are presented elsewhere (McNair, et al. 1992, Bergin, et al., 1995). Eulerian model results can also be compared across different parts of the modeled domain, which have varying VOC to NO x ratios because of pollutant emissions and transport, as well as variation in incident radiation caused by cloud cover (Bergin et al, 1995). In an initial application of a three-dimensional model for reactivity analysis, McNair et al. (1992) used the CIT air quality model with a relatively highly lumped chemical mechanism, the Lurmann, Carter, Coyner (1987) mechanism (LCC), to quantify the reactivity of 11 8

162 individual and lumped VOCs. This study allowed comparison with single-cell model reactivity studies by others, and also between different metrics of ozone impact, including how the species impact the peak ozone as well as ozone exposure. The results showed that the MIR reactivities did not perform well in predicting peak ozone sensitivities for the model species, but performed reasonably well in predicting effects of model species on integrated ozone exposures over the air quality standard. The MOIR scale did not compare as well as MIR to airshed model derived results for either the impacts on peak ozone or on ozone exposures over the air quality standard. The comparisons of McNair et al. (1992) are complicated somewhat by the fact that the study utilized the LCC chemical mechanism, which does not correspond directly with SAPRC-90 mechanism species used in calculating the MIR and MOIR scales. However, agreement between the MIR scale and the McNair et al. (1992) ozone exposure predictions is remarkably good considering the difference in the mechanisms, models, and ozone impact quantification techniques employed. It was noted in this study that much of the variability found could be ascribed to using a single species (CO in this case) for normalization, which is discussed further below. Subsequent to the study of McNair et al. (1992), the SAPRC-90 mechanism was implemented in the CIT model (referred to as the CIT-S90) by Bergin et al. for more direct comparison with the MIR and MOIR reactivity scales (Bergin et al, 1995, 1998). Here, reactivities are normalized to a mixture of VOCs representative of exhaust emissions, as in the reactivity studies of Carter (1994) and Yang et al. (1995). Some differences were found which are believed to be due to multi-day pollutant carryover and cloud cover represented in the CIT model, which are not accounted for by box models. The CIT-S90 was also used to investigate effects of environmental variabilities and of chemical mechanism uncertainties on reactivities (discussed in the variability and uncertainty sections, respectively, below.) A more detailed comparison of the CIT-S90 study results and the MIR and MOIR are also presented below. One other three-dimensional model study of reactivity is that of Kahn et al. (1998a,b), who conducted a reactivity study on VOC solvents having a wide range of reactivities. The SAPRC-90 chemical mechanism was used for this study, assuming rate constants of similarly reacting compounds for those solvents for which chamber studies have not been performed. The solvents studied included m-xylene (the most reactive), parachlorobenzotriflouride (PCBTF, a halogenated aromatic which is the least reactive), benzotriflouride (BTF), acetone, ethanol and isobutane. These compounds not only have a wide range of reactivities, but also represent a number of different types of VOCs. Using a box model to quantify the MIR and MOIR reactivities, as seen in Figure 3, they found reasonable agreement between the normalized MIR and MOIR reactivities, though the absolute reactivities differed by a factor of two. These results, along with others (Carter, 1994a; Bergin et al., 1995, 1997; Russell et al., 1995), suggest that normalizing reactivity (i.e., using relative reactivities) removes much of the environmental variation, though species reactivities do vary in regions of low NO x availability or very high VOC levels. Further, the VOC reactivity assessment for the U.S.-Mexico border region shows that it is currently possible to conduct reactivity assessments regionally, particularly when using a tool such as DDM. A further study (Yang et al., 1998) used a three-dimensional model to analyze both the uncertainties in relative reactivities and the spatial variability in the southern California area. That study is further discussed in the applicable sections, below. 9

163 The above studies deal primarily with assessing the reactivities of individual VOC species. In addition, a series of studies have looked at quantifying the reactivity of source emissions, e.g., how the reactivity of a mixture of VOCs emitted by a source might be affected by the methods employed, the location, and the metric of interest. A sampling of such studies is listed in Table 3. VARIABILITY IN REACTIVITY SCALES One of the stronger debates on the use of reactivity quantification for determining the potential impact of VOCs on ozone is that the absolute amount of ozone formed from a given quantity of VOC is heavily dependent on the local ambient conditions, including the meteorology (wind speed, temperature, mixing height and humidity), pollutant transport (the residence time of emissions in an urban area), distribution of emissions sources (e.g., proportion of biogenic, mobile source, and other emissions), and background pollutant concentrations (e.g., the VOC/NO x ratio and the absolute levels of VOCs and NO x ). This dependence on variable conditions was discussed above when presenting various reactivity scales and experimental results. One effect of variable conditions is that, in the extreme, a compound can go from being fairly reactive under certain conditions to having a negative reactivity under others (e.g. toluene). This dependence may make the use of generalized reactivity weighting and the development of reactivity-based control strategies problematic. However, one should also recognize that for most of the organics, those that are highly reactive relative to the other VOCs under one set of conditions remain highly reactive under other conditions. Likewise, the less reactive VOCs remain less reactive. Compounds that vary widely, such as toluene, are the exception rather than the rule. As discussed above, this variation can be reduced by the use of normalizing and relative ranking in the reactivity quantification of VOCs. For example, in 1990, CARB (1991, 1993) adopted the LEV/CF regulations, where vehicle exhaust emissions standards used Reactivity Adjustment Factors (RAFs), which are ratios of reactivities of the alternatively fueled to conventionally fueled exhausts, to account for the reactivity differences of these exhausts. The impact of environmental condition variability on RAFs was investigated by Russell et al. (1995), who calculated both the absolute reactivities and the RAFs of exhaust from vehicles operated on six fuels along each of the 39 trajectories used in developing the MIRs. The results are shown by box plots in Figure 4. The variation in absolute ozone forming potentials across cities is substantial. However, in the case of the RAFs, where the reactivities of the exhausts are normalized by the reactivity of standard gasoline exhaust, variation among cities is sharply diminished (Figure 4B). Again, it is the relative ozone impact that is of greatest concern. Such a marked decrease may not be found for source types emitting fewer compounds. A second issue in the analysis of variability in reactivity with environmental conditions is the effect of NO x and VOC background concentrations. The MIR scale was derived using conditions relatively high in NO x, as might be experienced in areas with a high density of NO x emissions (e.g., areas highly impacted by traffic or local industries with significant combustion sources). 10

164 MOIR conditions occur at lower NO x levels, but the ROG/NO x ratios are still lower than what might be found in rural areas. Reactivity simulation conditions used by Derwent and co-workers (1991) have even less NO x, which represents conditions where VOC controls and reactivity weighting are relatively ineffective (as discussed below). So the question arises as to how well a measure of reactivity quantification can represent many areas, given the possible range of environmental conditions under which ozone formation occurs. This was, in part, addressed above when comparing the MIR, MOIR, and POCP scales, and is addressed further below. The impact of environmental conditions on reactivity should be discussed at two levels. First, how it affects the reactivity of individual VOCs; and second, how it would likely affect the reactivity of emissions from a source whose composition is made up of a large number of VOCs. As suggested above, the absolute amount of ozone formed from any VOC is highly dependent on the environmental conditions. In an area already rich in VOCs (i.e., a NO x -limited regime), the small addition of an individual VOC has a lower impact than if that same increment of VOC emissions occurs in an area rich in NO x (where ozone formation is VOC-limited). As shown by Carter (1994), the average absolute reactivity of a suite of VOCs using the MIR scale is about twice that when using the MOIR scale. Further, there are those few compounds that can go from having relatively high reactivities to low or negative reactivities. This would appear to inhibit the use of reactivity weighting in regulatory applications. An interesting exercise that addresses the impact of environmental variabilities is a comparison between trajectory model results and three-dimensional model results. By their nature, the three-dimensional models cover domains with a wide range of environmental conditions, going from NO x -rich conditions in urban centers to VOC-rich conditions downwind. Further, they can follow the transport of pollutants over long distances. In the Bergin et al. (1995, 1998) modeling studies described previously, the spatially and temporally resolved ozone impacts were used to calculate impacts on the peak ozone, the potential population-weighted ozone exposure, and the spatial-weighted ozone exposure. From those calculations, the corresponding compound reactivities were quantified, normalized to the reactivity of a mixture of VOCs (so the results are relative reactivities.) As shown in Figure 5, the results from the MIR and MOIR box model calculations (Carter, 1994a), conducted for 39 cities, agreed well with related metric results from the airshed calculations for the Los Angeles, California area (as described further below). In interpreting the results of the comparison between the two modeling approaches, and the differences found between the three metrics defined for the airshed model results, it is important to understand the ozone and population patterns in the region. The peak ozone is found in the eastern basin, in an area with relatively little NO x, and thus has a high VOC/NO x ratio. On the other hand, the population is concentrated more in the western basin, in areas with more dense emissions, and in particular NO x -rich mobile source emissions, and thus having a low VOC/NO x ratio. Also, the peak ozone is found downwind of the urban area, after the pollutants have had a chance to age, again in contrast to the more densely populated regions that experience fresh emissions. Further, the meteorology (e.g., temperatures and mixing heights) in the two portions of the basin is different. Because of these differences, contrasting the populationweighted ozone impact with the peak ozone impact can help capture the level of difference found 11

165 from environmental variability. The spatial-exposure metric is expected to give results with characteristics of each of the other metrics. As shown in Figure 5, the airshed model-derived spatial and population density weighted results behave similarly to MIRs. The greatest differences are found for formaldehyde and other compounds whose reactivities are highly dependent on photolytic reactions. This may be explained by the use of a reduced photolysis rate in the airshed modeling to account for the observed cloud cover. The box model used clear sky conditions. The reductions in the reactivities are consistent with the sensitivity to the rate constants for the photolytic reactions (Yang et al, 1995), as addressed in the uncertainty section below. In general, airshed model results for Los Angeles agree well with MIRs, and further show that individual organics have very different ozone impacts. Such a study has not been conducted for other regions. Russell et al. (1995), Bergin et al. (1995), and Kahn et al. (1998) compared several reactivity metrics for the box and grid models for some two dozen plus compounds including some half dozen aromatics. In general, aromatics turn out to be significantly less reactive or negative, on a relative basis, for some metrics. Since the EPA is moving to an 8 hour ozone standard, this should be the primary metric of choice. It appears as if additional work is needed to look at aromatics under several different conditions. Eastern transport conditions should also be examined in a multiday scenario. A number of the aromatics, important for industrial emissions, may be different on a relative basis. To further compare the trajectory and airshed model results, regression analysis was performed between the box model reactivities and the airshed reactivities. As shown in Table 4, Carter s MIR scale corresponded well with the population exposure-based reactivities, and the MOIR scale agreed well with the CIT-S90 peak-ozone sensitivity. In these two cases, the slope of the regression line is virtually 1 (showing little bias as reactivities increase), and the correlation is high. The CIT-S90 spatial exposure metric correlates well with both the MIR and MOIR scales, but shows some bias in the comparison with the MOIR scale, indicating that the spatial exposure metric finds the less reactive compounds to be relatively more reactive than does the MOIR scale. As seen from Figure 5, there are significant similarities between the CIT-S90 metrics as well, though some differences are evident. Differences were quantified by calculating the normalized bias (a value of 1 would indicate a 100% bias) and standard deviation between the scales (Table 5). These differences between potential metrics for reactivity quantification within a modeling study also introduce variability. A similar issue in regards to the role that environmental variability plays in reactivity quantification is how various meteorological characteristics can affect reactivities. Russell et al. (1995) studied the variability in reactivities as found using the results of the box model of Carter under differing conditions, a similar box model (Yang et al, 1995), and a three-dimensional model (Bergin et al, 1995). First, using just the results of the box model calculations of Carter (1994), they quantified the inter-city variability in the absolute species reactivities along the 39 trajectories, and the inter-city variability in the relative reactivities of the individual VOCs along those same trajectories. Normalized MIRs were calculated by dividing each species city-specific 12

166 MIR by the geometric mean reactivity of all the species reactivities for that city, and multiplying by the geometric mean reactivity of the 39-city average MIRs. This alleviates the problem that VOCs are generally less reactive, in an absolute sense, in one city versus another. A sample of their results is given below in Table 6. As seen, the variability in the relative reactivity is significantly reduced between the different trajectories when the relative reactivities are used. Use of relative reactivity generally reduced variability by almost a factor of two, from about 20% to 12%. More recently, Kahn et al. (1998a,b), Qi et al. (1998) and Yang et al. (1988) also studied spatial variability in VOC reactivity. Kahn et al., (1998a) approached this by applying a three dimensional model to different domains (Los Angeles, Mexico City and Switzerland) that experience very different levels of ozone precursors. Los Angeles has relatively high levels of NOx in most of the domain, and moderate levels of VOC. Switzerland has lower NOx and VOC levels. Mexico City has high NOx levels, and very high VOC levels. While the Los Angeles and Switzerland reactivity results were similar, the Mexico City reactivities varied, suggesting that having a large amount of VOC has a greater impact on compound reactivity than does the level of NO x. This was further indicated in the Yang et al., (1998a) study, discussed above, which found that the sensitivity of the relative reactivities to the uncertainties in the VOC emission level was greater than to those in the NO x level. Yang et al. (1998), also plotted the spatial variation in the VOC reactivities in the Los Angeles basin. Not surprisingly, the relative reactivities do vary (Figure 6). Formaldehyde has its greatest relative reactivity in the source regions since it reacts quickly and is a strong radical source, further increasing VOC oxidation in a NOx-rich environment. The slower reacting compounds (e.g., pentane) have a proportionally higher relative reactivity downwind. VOCs that can become significant NO x sinks (e.g., toluene) have the greatest variation in reactivity, going from being positive in the source region to negative far downwind. This spatial variability suggests that it is important to define the endpoint of interest: e.g., peak ozone, ozone exposure, and to calculate the corresponding relative reactivity. Kahn et al. (1999b) applied the CIT model using the 3-D Direct Decoupled Method (DDM3D) to the U.S.-Mexico border region. This study looked at both the use of direct sensitivity analysis for reactivity assessment, as well as spatial variation in relative reactivity. The use of DDM3D produced similar results to the brute force approach with less computational noise and effort. In the study by Yang et al., (1998), Monte Carlo analysis was applied with the CIT/DDM3D model, to quantify how influential rate constants, VOC and NOx emissions and meteorology (i.e., mixing height) impacted the relative reactivity of 11 VOCs. Figure 6 also shows the spatial variability in the coefficient of variation of the HCHO relative reactivity. The uncertainty is relatively uniform at about 35% over the middle of the domain. In addition, multiple linear regression was used to find how individual species reactivities depended upon model parameters (e.g., rate constants) and inputs (emissions of VOC and NOx, and mixing heights). For example, the relative reactivities for the ozone exposure to HCHO and pentane were found to be: 13

167 RRHCHO = FNO + hν FVOC FHCHO+ hν 0.43FPAN decomp 11. F F O + hν O D NO2 + HO 3 2 RRpen tane = F 0 02F F O h O D VOC e AFG h ν 2+ ν FNO + HO FAAR2+ HO FOSD+ H O 2 2 The independent variables, the F s, represent the ratio of the variable of interest to the default value, the VOC e subscript represents the amount that the total VOC emissions were changed, and the other subscripts (e.g., HCHO+hv) represent the corresponding reaction rates. Only the six most important parameters are given. The correlation coefficients in both cases are about Both regression and uncertainty analyses suggest that the most influential variables were the formaldehyde photolysis rate, ozone photolysis to O 1 D, the NO2-HO reaction and the VOC emission rate. On the other hand, the mixing height and NO x emission rate had relatively less impact. It should be noted that these results may be specific to the domain modeled, and should not be generalized to being true for other areas. In fact, it might be suspected that the sensitivity to the VOC emissions rate may be particular to area like Los Angeles, and that would not be the case for a location like Mexico City (high VOC area) or much of the less-urbanized United States (low NO x ), or areas with a larger influence of ozone precursors being transported into the region. This is an obvious area for further exploration. e CHEMICAL MECHANISM UNCERTAINTY As discussed in the Atmospheric Chemistry section, the chemical mechanisms used in the models to calculate the reactivities of the VOCs have significant uncertainties, which causes a corresponding uncertainty in results of reactivity assessments Measurement errors in laboratory kinetic and product studies contribute to a minimum level of uncertainty in the mechanisms of even the best studied VOCs, and the reactions of many of the organic compounds emitted into urban atmospheres have never been studied in controlled experiments. Their representation in chemical mechanisms is based on analogy to compounds of similar structure, creating added uncertainty, which is difficult to quantify. At issue is the extent to which the uncertainties in the chemistry impact the calculation of the reactivities for organic compounds. One way to assess the effects of chemical mechanism uncertainty is to compare reactivity predictions using different mechanisms which represent the same state of the art but which incorporate differing assumptions concerning unknown areas of the chemistry and differing condensation approaches. As discussed above, the SAPRC-90 mechanism was used for calculation of the MIR, MOIR and other reactivity scales because of the number of VOCs it can explicitly represent. The RADM-II mechanism employs assumptions similar to SAPRC-90 concerning uncertain portions of the aromatics and other mechanisms, and would be expected to give similar reactivities for the species that the condensed mechanisms are designed to represent. However, this may not be the case for the Carbon Bond IV (CB4) mechanism, which employs differing assumptions concerning some of the uncertainties in the aromatics mechanisms, and 14

168 uses different methods for treating alkane and alkene reactions (Gery et al., 1988). Table 7 shows a comparison of MIR and MOIR (relative to the total base case emissions) calculated with the SAPRC-90 and with a recent version of the CB4 (with minor updates concerning peroxy radical reactions that do not significantly affect ozone predictions (Yarwood, 1994; Carter, 1994b). Other than the mechanism, the scenarios and the calculation methodology are the same (Carter, 1994b). Note that the only compounds shown are those that are either represented explicitly in CB4 or are represented by model species developed based on mechanisms of very similar compounds, so the differences reflect primarily differences in representation of the actual reactions, rather than condensation effects. Note also that the SAPRC-90 and the CB4 mechanisms were developed around the same time, so they are both based on approximately the same data base of kinetic, mechanistic, and environmental chamber results. The most conspicuous difference on Table 7 is for toluene, for which the developers of the CB4 added a speculative reaction so model simulations could accurately predict the relatively low maximum ozone yields in some toluene-no x outdoor chamber experiments (Gery et al., 1988). This reaction is not included in the SAPRC-90 mechanism, nor is it in the CB4 mechanism for xylenes. This causes somewhat lower MIR reactivities for toluene and causes toluene to be negatively reactive at the lower NO x levels where maximum ozone formation occurs. (The SAPRC-90 mechanism also predicts that toluene becomes negatively reactive at low NO x levels, but the NO x levels must be much lower than is the case with CB4.) In the case of xylenes, where the CB4 lacks this speculative reaction, the MIR and MOIR relative reactivities agree quite well. The somewhat higher CB4 formaldehyde reactivity is believed to be caused primarily by a greater sensitivity of the CB4 mechanism to radical input processes, rather than by differences in the formaldehyde mechanism itself. Another way to assess the effects of mechanism uncertainty is to examine changes in calculated reactivities that result when mechanisms are updated. Since SAPRC-90 was developed there has been new kinetic and mechanistic data concerning a number of relevant reactions, and environmental chamber studies of a number of compounds. The effects of two rounds of updates to the SAPRC mechanism on MIR s for selected compounds are shown on Table 8. The SAPRC-97 mechanism is an updated version of SAPRC-90 which incorporates significant updates in PAN kinetics, major modifications to alkene and aromatic reactions and mechanisms for other individual VOCs resulting from new mechanistic and environmental chamber data (Carter et al, 1997). The SAPRC-98 represents a complete update to the rate constants of the inorganic and common product reactions, somewhat more detailed representation of reactive organic products, revised estimation methods for OH + alkanes and oxygenates, and re-evaluations against the chamber data (see Carter, 1998 and the Atmospheric Chemistry section.) Table 8 shows that updates to the base mechanism can cause a relative MIR change on the order of ~15% even for explicitly represented compounds whose mechanisms were not changed. In the case of going from SAPRC-90 to SAPRC-97, the changes for some of the aromatic isomers is due to new environmental chamber data, and the changes for acetylene and methyl isobutyrate are due to changes in the estimated, unverified mechanisms. (Differences between SAPRC-90 and SAPRC-97 are discussed further below.) In the case of going from SAPRC-97 to SAPRC-98, the relatively large increases in MIR for the higher alkanes, 2-(2-15

169 ethoxyethoxy), ethanol and other compounds is due largely to the use of a more reactive model species to represent more reactive non-aldehyde oxygenated products, the changes for acetylene and methyl isobutyrate are due to mechanism revisions resulting from new chamber data, and in the case of the higher ketones the change is due to representing these compounds explicitly rather than by MEK. It can be seen that changes of up to a factor of 3 have occurred for compounds with unverified mechanisms. More systematic studies of the effects of mechanism uncertainties have been carried out using airshed and box models to explore to what degree uncertainties in chemical rate parameters affect the calculated compound reactivities (Yang et al., 1995; 1996a,b; Bergin et al., 1996; 1997; Wang and Milford, 1998). Figure 7 compares results from box model uncertainty analysis studies conducted by Yang et al. (1996b) using the SAPRC-90 chemical mechanism and by Wang and Milford (1998) using the SAPRC-97 mechanism. Both studies used Monte Carlo analysis with Latin Hypercube Sampling to calculate the uncertainties in MIRs and MOIRs. Computational requirements were reduced by using a single set of trajectory conditions designed by Carter (1994a) to give incremental reactivities close to the average from the 39 trajectories used in developing his scales. Uncertainty estimates for parameters in the SAPRC-90 mechanism were compiled by Stockwell et al. (1994), largely from concurrent reviews of kinetic data (DeMore et al., 1990; Atkinson et al., 1989). For the SAPRC-97 mechanism, the uncertainty estimates were updated using more recent reviews (DeMore et al., 1994; 1997). In addition, Wang and Milford (1998) rectified a shortcoming of the earlier study by developing original estimates of uncertainties in aromatics oxidation parameters. Aromatics oxidation chemistry is highly uncertain, and the parameters used in chemical mechanisms are estimated by fitting ozone production due to aromatics oxidation in environmental chamber experiments. Accordingly, the uncertainty estimates that Yang et al. (1996a) used for these parameters were very large and highly subjective. Wang and Milford (1998) applied stochastic parameter estimation techniques to the aromatics database from the University of California at Riverside environmental chambers, to objectively assess the uncertainty in the aromatics oxidation parameters. As shown in Figure 7, mean MIR estimates calculated with SAPRC-97 are generally higher than those calculated with SAPRC-90, reflecting revisions to the mechanism. One exception is the MIR for 1,2,4-trimethylbenzene, which has been adjusted downward based on recent chamber experiments. Yang et al.'s (1996) uncertainty estimates for MIRs ranged from about 30% to 50% of the mean MIR values, for most compounds. Uncertainty estimates for most aromatic compounds fell at the upper end of that range. The recent uncertainty estimates for MIRs made by Wang and Milford (1998) are somewhat lower, ranging from 25% to 35% in most cases. Reduced uncertainty estimates for aromatic compound reactivities suggest that the chamber data effectively constrain the uncertainty in these values. Both studies indicate that uncertainties in MOIRs are somewhat higher than those for MIRs. For SAPRC-97, Wang et al. estimate that uncertainties in MOIRs range from about 30% to 50% for most compounds. As noted by Yang et al. (1995, 1996b) uncertainties in many rate parameters have similar effects on the reactivities of various compounds, so the resulting incremental reactivities are strongly correlated. For example, an increase in the photolysis rate for NO2 increases the reactivity of most species by about the same proportion. Thus, the relative reactivity of one 16

170 species compared to another is not affected as much as the absolute incremental reactivities by chemical parameter uncertainties. Bergin et al. (1997) extended the box-model rate constant uncertainty studies to a threedimensional model uncertainty study. After the most influential rate parameters were identified by Yang et al. through Monte Carlo simulations, described above, those values in the CIT-S90 model were varied by twice the estimated uncertainty, and the compound reactivity simulations were then recalculated. Results of one metric studied (spatial exposure) are shown in Figure 8. This analysis, again, found that relative reactivities have relatively low sensitivities to rate constant uncertainties. The implication of this result is further demonstrated by considering uncertainties in source reactivity quantification and RAFs. Further analysis of uncertainties using a three-dimensional model was conducted by Yang et al., (1998). As shown in Table 9, the uncertainties in the local relative reactivity were about 30%, accounting for uncertainties in 14 of the most influential rate constants, VOC and NOx emissions, and mixing height. Table 9 rank orders the sources of uncertainty. They found that the major source of uncertainty in quantifying most of the VOC reactivities were the NO2+HO reaction rate and the VOC emissions rate. REFERENCES Altshuller, A. P. and Bufalini, J. J. (1971). Environ. Sci. Technol, 5: 39. Andersson-Skold, Y.; Grenfelt, P.; and Pleije, K. (1992). J. Air Waste Mgmt. Assoc., 42: AQIRP (1993). Reactivity Estimates for Reformulated Gasolines and Methanol/Gasoline Blends in Prototype Flexible/Variable Fuel Vehicles, Technical Bulletin No. 7, Auto/Oil Air Quality Improvement Research Program. Available from the Coordinating Research Council, Atlanta, GA. Atkinson, R. (1987). Int. J. Chem. Kinet., 19: Atkinson, R. (1990). Atmos. Environ., 24A:1 24 Atkinson, R. (1989). J. Phys. Chem. Ref. Data, Monograph No 1.Atkinson, R. (1994). J. Phys. Chem. Ref. Data, Monograph No. 2. Bergin, M.S.; Russell, A.G.; and Milford, J.B. (1995). Environ. Sci. & Tech., 29(12): Bergin, M.S.; Russell, A.G.; and Milford, J.B. (1998). Environ. Sci. & Tech., 32(5): Bowman, F.M. and Seinfeld, J.H. (1994a). J. Geophys. Res., 99: Bowman, F. M. and Seinfeld, J.H. (1994b). Atmos. Environ., 28: Bowman, F. M. and Seinfeld, J.H. (1995). Prog. in Energy and Comb. Sci. 21:

171 Bufalini, J. J.; Walter, T.A.; and Bufalini, M.M. (1977). Environ. Sci. Technol., 11: CARB. (1989). Definition of a Low-Emission Motor Vehicle in Compliance with the Mandates of Health and Safety Code Section (Assembly Bill 234, Leonard, 1987), Report by Mobile Sources Division, California Air Resources Board, El Monte, CA. CARB. (1990). Proposed Regulations for Low-Emission Vehicles and Clean Fuels Staff Report and Technical Support Document, Sacramento, CA, August 13. See also Appendix VIII of California Exhaust Emission Standards and Test Procedures for 1988 and Subsequent Model Passenger Cars, Light Duty Trucks and Medium Duty Vehicles, as last amended September 22, Incorporated by reference in Section (k) of Title 13, California Code of Regulations. CARB. (1991). Proposed Reactivity Adjustment Factors for Transitional Low-Emissions Vehicles Staff Report and Technical Support Document, Sacramento, CA, September 27. Andersson-Skold, Y.; Grenfelt, P.; and Pleije, K. (1992). J. Air Waste Mgmt. Assoc., 42: Carter, W.P.L. (1990). Atmos. Environ., 24A: Carter, W.P.L. (1991). Development of Ozone Reactivity Scales for Volatile Organic Compounds, EPA 600/ U.S. Environmental Protection Agency, Research Triangle Park, NC (August). Carter, W.P.L. (1994a). J. Air and Waste Mgmt. Assoc., 44: Carter, W.P.L. (1994b). Calculation of Reactivity Scales Using an Updated Carbon Bond IV Mechanism, Report prepared for Systems Applications International under funding from the Auto/Oil Air Quality Improvement Research Program. Carter, W.P.L. (1995). Atmos. Environ., 29: Carter, W.P.L. (1998): Updated Maximum Incremental Reactivity Scale for Regulatory Applications, Preliminary Report to California Air Resources BoardContract No See Carter, W.P.L.; Atkinson, R.; Winer, A.M.; and Pitts, J.N. Jr. (1982). Int. J. Chem. Kinet., 14:1071. Carter, W.P.L. and Atkinson, R. (1989). Environ. Sci. and Technol., 23:864. Carter, W.P.L. and Lurmann, F.W. (1990). Evaluation of the RADM Gas-Phase Chemical Mechanism, Final Report, EPA-600/ Carter, W.P.L. and Lurmann, F.W. (1991). Atmos. Environ., 25A:

172 Carter, W.P.L.; Pierce, J.A.; Malkina, L; Luo, D.; and Long, W.D. (1993). Environmental Chamber Studies of Maximum Incremental Reactivities of Volatile Organic Compounds, Report to Coordinating Research Council, Project No. ME-9, California Air Resources Board Contract No. A ; South Coast Air Quality Management District Contract No. C91323, United States Environmental Protection Agency Cooperative Agreement No. CR , University Corporation for Atmospheric Research Contract No , and Dow Corning Corporation (April 1). Carter, W.P.L.; Luo, D.; Malkina, I.L.; and Pierce, J.A. (1995a). Environmental Chamber Studies of Atmospheric Reactivities of Volatile Organic Compounds. Effects of Varying ROG Surrogate and NOx. Final report to Coordinating Research Council, Inc., Project ME-9, California Air Resources Board, Contract A , and South Coast Air Quality Management District, Contract C91323 (March 24). Carter, W.P.L.; Pierce, J.A.; Luo, D.; and Malkina, I.L. (1995b). Atmos. Environ., 29: Carter, W. P. L., D. Luo, and I. L. Malkina (1997): "Environmental Chamber Studies for Development of an Updated Photochemical Mechanism for VOC Reactivity Assessment," final report to California Air Resources Board Contract , Coordinating Research Council Project M-9, and National Renewable Energy Laboratory Contract ZF November 26. Chameides, W.L.; Fehsenfeld, F.; Rodgers, M.O.; Cardelino, C.; Martinez, J.; Parrish, D.; Chang, T.Y. and Rudy, S.J. (1990). Atmos. Environ. 24A(9): Chang, T.Y.; Rudy, S.J.; Kuntasal, G.; Gorse, R.A., Jr. (1989). Atmos. Environ. 23(8): Darnall, K.R.; Lloyd, A.C.; Winer, A.M.; Pitts, J.N. Jr. (1976). Environ. Sci. Technol., 10:692. Derwent, R.G. and A.M. Hov. (1979). Computer Modeling Studies of Photochemical Air Pollution Formation in North West Europe, AERE R 9434, Harwell Laboratory, Oxfordshire, England. Derwent, R.G. and Jenkin, M.E. (1991). Atmos. Environ., 25(A): Dimitriades, B. (1996): "Scientific Basis for the VOC Reactivity Issues Raised by Section 183(e) of the Clean Air Act Amendments of 1990," J. Air Waste Manage. Assoc. 46, Dodge, M.C. (1977) "Combined Use of Modeling Techniques and Smog Chamber Data to Drive Ozone-Precursor Relationships" in Proceedings, International Conference on Photochemical Oxidant and Its Control, Research Triangle Park, North Carolina, EPA- 600/ a. Carter, W.P.L. (1990). Atmos. Environ., 24A:

173 Dodge, M.C. (1984). Atmos. Environ., 18:1657. In B. Dimitriades, ed., International Conference on Photochemical Oxidant Pollution and Its Control Proceedings: Volume II, September Research Triangle Park, NC. Gery, M.W. and Crouse, R.R. (1989). User's Guide for Executing OZIPR. Atmospheric Research and Exposure Assessment Laboratory, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina. Jeffries, H.E. and Sexton, K.G. (1993). The Relative Ozone Forming Potential of Methanol Fueled Vehicle Emissions and Gasoline-Fueled Vehicle Emissions in Outdoor Smog Chambers, Draft Final Report to the Coordinating Research Council, Project No. ME-1 (September). Jeffries, H.E. and Crouse, R. (1991). Scientific and Technical Issues Related to the Application of Incremental Reactivity. Part H: Explaining Mechanism Differences, Report prepared for Western States Petroleum Association, Glendale, CA (October). Jiang, W.; Singleton, D.L.; Hedley, M.; and McLaren, R. (1996) Atmos. Environ. 31(4): Joshi, S.B.; Dodge, M.C.; and Bufalini, J.J. (1982). Atmos. Environ., 16: Kahn, M.; Yang, YJ; and Russell, A.G. (1997). "Reactivities of Organic Solvents: Comparison between Urban and Regional Domains," Paper No. WA-97-!1540. Presented at the A&WMA Annual Meeting, Toronto, Canada. June. Air & Waste Management Association, Pittsburgh, PA. Kahn, M., Yang, YJ and Russell, A.G.., (1999 a) Photochemical reactivities of common solvents: comparison between urban and regional domains, Atmos. Env. (in press) Kahn, M.; Qi, L.; Yang, YJ and Russell, A.G.(1999b) Regional assessment of realtive reactivities using a decoupled, direct method for sensitivity analysis (to be submitted) Laity, J.L.; Burstain, F.G.; and Appel, B.R. (1973). In R.W. Tess, ed., Solvents Theory and Practice, Adv. Chem. Series, 124:95. Lonneman, W.; Lawson, D.R.; Rasmussen, R.A.; Zimmerman, P.; Greenburg, J.; Middleton, P.; and Wang, T. (1992). J. Geophys. Res., 97:6037. Lurmann, F.W.; Carter, W.P.L.; and Coyner, R.A. (1987). A Surrogate Species Chemical Reaction Mechanism for Urban-Scale Air Quality Simulation Models. Volume I - Adaptation of the Mechanism, EPA-600/ a. McNair, L.; Russell, A.; and Odman, M.T. (1992). J. Air Waste Mgmt. Assoc., 42:174. Odman et al., 1994 (URM model)russell, A.; Milford, J.; Bergin, M.S.; McBride, S.; McNair, L.; Yang, Y.; Stockwell, W.R.; and Croes, B. (1995). Science 269: Russell, A.G.; St. Pierre, D.; and Milford, J.B. (1990). Science 247:

174 Russell, A.G.; Milford, J.; Bergin, M.; McNair, L.; Yang, YJ.; and Stockwell, W. (1995) Science, 269: Trijonis, J.C. and Arledge, K.W. (1976). Utility of Reactivity Criteria in Organic Emission Control Strategies: Application to the Los Angeles Atmosphere, EPA/600/3-78/019; TRW/Environmental Services: Redondo Beach, CA. Weir, B.R.; Rosenbaum, A.S.; Gardner, L.A.; Whitten, G.Z.; and Carter, W. (1988). Architectural Coatings in the South Coast Air Basin: Survey, Reactivity, and Toxicity Evaluation, Final Report to the South Coast Management District, SYSAPP-88/137, Systems Applications, Inc., San Rafael, CA (December). Wilson, K.W. and Doyle, G.J. (1970). Investigation of Photochemical Reactivities of Organic Solvents, Final Report, SRI Project PSU-8029, Stanford Research Institute, Irvine, CA (September). Yang, Y.-J.; Das, M.; Milford, J.B.; Bergin, M.S.; Russell, A.G.; and Stockwell, W.R. (1994). Quantification of Organic Compound Reactivities and Effects of Uncertainties in Rate Parameters. An Integrated Approach Using Formal Sensitivity and Uncertainty Analysis and Three Dimensional Air Quality Modeling. Report prepared for the Auto/Oil Air Quality Improvement Research Program (August). Yang, Y-J.; Stockwell, W.R.; and Milford, J.B. (1995). Environ. Sci. & Tech., 29(5): Yang, YJ.; Khan, M.; Wilkinson, J.; and Russell, A.G. (1999) Spatial uncertainty assessment of relative reactivities (in preparation) Yarwood, G. (1994): System Applications, Inc., San Rafael, CA, personal communication with W.P.L. Carter. 21

175 TABLES Table 1: Summary of compound reactivity modeling studies Reference Model Type Mechanism Application Derwent and Jenkin (1991) trajectory Harwell Two-layer multi-day trajectory simulations of reactivity. Referred to as POCP scales. McNair et al. (1992) threedimensional (CIT) LCC Calculation of 3 reactivity scales for 11 lumped compounds. Simulations were performed for a threeday period in the Los Angeles area (the SCAQS episode). Carter (1994a) trajectory SAPRC-90 Development of 18 reactivity scales (including the MIR and MOIR) for 117 compounds. Results are the average of 39 trajectory simulations for 10-hour periods. Yang et al. (1994) trajectory and three-d (CIT) SAPRC-90 Review of rate constant uncertainties and also portions of (Yang et al, 1995; Bergin et al, 1995, 1998). Report. Yang et al. (1995) Bergin et al. (1995) Jiang et al. (1996) Bergin et al. (1997) Kahn et al. (1999a) trajectory SAPRC-90 Rate constant uncertainty calculations for the reactivities of 26 compounds under MIR- and MOIRtype conditions. One averaged trajectory was used rather than the 39 used in the Carter MIR and MOIR calculations. threedimensional (CIT) SAPRC-90 Calculation of 3 reactivity scales for 27 compounds. Simulations were performed for the SCAQS episode. trajectory SAPRC-90 Calculation of the contributions of eighteen compounds to ozone concentrations in the Lower Fraser Valley. threedimensional (CIT) SAPRC-90 SAPRC-90 Rate constant uncertainty calculations for the scales and compounds in the Bergin et al. (1995) study above. Calculation of solvent reactivities using box and airshed modeling in multiple domains. Carter (1998) trajectory SAPRC-98 Update of Carter (1994) Kahn et al., (1999b) Yang et al. (1999) trajectory and threedimensional threedimensional threedimensional SAPRC-90 (with updates) SAPRC-90 (with updates to 97) Regional uncertainty analysis and use of direct sensitivity method Spatial uncertainty analysis of reactivities 22

176 Table 2. Summary of major characteristics of Carter (1994) reactivity scales. Scale Type of Scenarios used Derivation of scale from individual scenario reactivities Ozone quantifica tion Reflects effect of VOC on: Maximum Incremental Reactivity (MIR) Low ROG/NO x conditions where O 3 is most sensitive to VOC changes Averages of incremental reactivities in the MIR scenarios Maximum ozone Ozone formation rates Maximum Ozone Incremental Reactivity (MOIR) Moderate ROG/NO x conditions where highest O 3 yields are formed Averages of incremental reactivities in the MOIR scenarios Maximum ozone Ultimate ozone yield Equal Benefit Incremental Reactivity (EBIR) Higher ROG/NO x conditions where VOC and NO x control are equally effective in reducing O 3 Averages of incremental reactivities in the EBIR scenarios Maximum ozone Ultimate ozone yield Base-Case Average Ratio: O 3 Yield Base case conditions (ROG/NO x conditions are as observed for the individual scenarios) Averages of incremental reactivities in the base case scenarios Maximum ozone Ultimate ozone yield Base-Case Least Squares Error: O 3 yield. Base-Case Minimizes change in ozone if a "null test" substitution were made using the scale [a] Maximum ozone Depends on the variability of scenario conditions [b] Base-Case Average Ratio: Integrated O 3 Base-Case Averages of incremental reactivities in the base case scenarios Integrated ozone Ozone formation rate and ultimate yield Base-Case Least Squares Error Integrated O 3 Base-Case Same as base case least squares error O 3 yield. [b] Integrated ozone Ozone formation rate [a] [b] A "null test" substitution based on a reactivity scale consists of substituting VOC emissions such that the scale predicts there would be no change in ozone. Depends on effect on O 3 formation rate if scenarios are highly varied in ROG/NO x conditions. 23

177 Table 3. Summary of source emissions reactivity modeling studies Reference Model Type Mechanism Application Trijonis and Arledge (1976) calculated (not modeled) EPA Smog Chamber Data Estimated major source reactivities for metropolitan Los Angeles. Chang et al. (1989) trajectory LCC Methanol fuel vehicle impacts with respect to conventionally fueled vehicles. Russell et al. (1990) threedimensional (CIT) LCC Potential methanol fuel vehicle impacts for the SCAQS episode (compared to equal mass emissions from conventional vehicles). McNair et al. (1994) (100) threedimensional (CIT) LCC Calculations of RAFs for 4 fuels. Simulations were performed for the SCAQS episode. Yang et al. (1996) (101, 102) trajectory SAPRC-90 Rate constant and exhaust composition uncertainty calculations for the RAFs from reformulated gasolines and methanol. Bergin et al. (1996) (103) trajectory and threedimensional (CIT) SAPRC-90 Report on box model study described above (101, 102) and a 3D study of the effects of rate constant and product yield uncertainties on predicted ozone impacts of 5 alternative fuel RAFs. Russell et al. (1995) (96) trajectory and threedimensional (CIT) SAPRC-90 Evaluation of combined results of most previous studies (82, 84, 85, 100, 101, 103). An economic analysis was also performed. Guthrie et al. (1996) (104) threedimensional (UAM) CB4 Modeling of potential impacts of the use of three alternative fuels (CNG, M85, and RFG) in two urban areas. Report. 24

178 Table 4. Regression results for airshed model exposure vs. MIR and airshed peak ozone vs. MOIR measures (Bergin et al., 1996, 1998) Comparison R 2 Slope Intercept Population Exposure to MIR Spatial Exposure to MIR Peak to MOIR Spatial Exposure to MOIR Table 5. Normalized bias and standard deviation between reactivity metrics calculated using the CIT-S90 airshed model (Bergin et al., 1996, 1998) Comparison Bias Standard Deviation Population to Spatial Exposure Peak Ozone to Spatial Exposure Peak Ozone to Population Exposure Table 6. Example MIRs and variations between locations (mean and standard deviation) (Russell et al., 1995) Compound Mean reactivities across 39 cities (Non-normalized/Normalized) Standard deviation (Non-normalized/Normalized) HCHO 7.2/ /0.58 Methanol 0.56/ /0.064 Ethane 0.25/ /0.045 Toluene 2.7/ /0.28 Pentene 6.2/ /

179 Table 7. Comparision of relative reactivities calculated using the SAPRC90 and the Carbon Bond 4 mechanisms. Compound Reactivity relative to base ROG mixture (mass basis) MIR MOIR S-90 CB4 Diff S-90 CB4 Diff n-hexane % % Ethene % % Propene % % trans-2-butene % % Toluene % % m-xylene % % Formaldehyde % % Acetaldehyde % % Methylethy ketone % % Methanol % % Ethanol % % 26

180 Table 8. Effect of SAPRC mechanism updates on calculated MIR's. Compound Reactivity relative to base ROG mixture (mass basis) SAPRC Version Change >97 90->98 97->98 Ethane % 15% 15% n-octane % 63% 84% n-pentadecane % 58% 71% 2,4-Dimethyl Heptane % 20% 15% Ethene % 9% 23% Propene % 7% 17% 1-Hexene % 15% 15% trans-2-butene % 19% 16% Isoprene % 2% 27% a-pinene % 24% 35% Toluene % 23% -16% p-xylene % -46% 58% m-xylene % 9% -19% 1,3,5-Trimethyl Benzene % -12% -16% 1,2,4-Trimethyl Benzene % -32% 45% Naphthalene % 109% 150% Acetylene % 94% 256% Ethanol % 15% 15% Propylene Glycol % 38% 11% Methyl t-butyl Ether % 72% 91% 2-(2-Ethoxyethoxy) EtOH % 139% 67% Methyl Isobutyrate % 72% -42% Formaldehyde % 3% 44% C3 Aldehydes % 3% 15% Acetone % -31% 2% C6 Ketones % 251% 304% 27

181 Table 9. Rank ordering and source attribution of uncertainty in relative reactivities for airshed modeling of Los Angeles. Parameter Relative Reactivity, Uncertainty and Contributions to Relative Reactivity Uncertainty Overall Relative Reactivity/ uncertainty/cov HCHO Pentane Average over 11 VOCs 5.76/1.05/ /0.051/0.12 COV average =0.19+/-0.09 Paramter Uncertainty Contributions NO2+hv O3+NO O3+hv-->O 1 D O 1 D+H2O O 1 D+M HO +NO HO2+NO RO2R+NO AFG2+hv CCO+NO CCO+NO PAN decomposition HCHO+hv AAR2+HO NOx Emissions VOC Emissions Mixing Height

182 FIGURES 8 Constant ozone isopleths NOx VOC Ozone (ppm) Constant VOC = 10 m.mol/m 2 MOR MIR [O3] d[o3]/d[voc] NOx (m.mol/m 2 ) d[o3]/d[voc] (ppm/m.mol/m 2 ) Ozone (ppm) Constant NOx = 2.5 m.mol/m 2 MIR [O3] d[o3]/d[voc] VOC (m.mol/m 2 ) d{o3]/d[voc] (ppm m.mol/m 2 ) 1 MOR, moximum ozone reactivity, also referred to as maximum ozone incremental reactivity (MOIR); MIR, maxium incremenmtal reactivity. One day maximum ozone concentrations calculated in a one-day box model simulation using the averaged conditions scenario of Carter (1994) and the SAPRC-93 chemical mechanism. Figure 1. Dependence of peak ozone and d[o3]/d[voc] on VOC and NOx 29

183 Figure 2. Comparison of MIR, MOIR, and POCP relative reactivities MIR MOIR POCP Ethane n-butane n-octane n-c12 Propene trans-2-butene Benzene Toluene m-xylene 1,2,4- Trime.benzene Formaldehyde Acetaldehyde Methyl Ethyl Ketone Methanol Ethanol 1 Incremental reactivities (ozone per gram) are shown relative to ethene = 100. MIR and MOIR reactivities from Carter (23). POCP reactivities are averages for various trajectories calculated by Derwent and Jenkin (81), with the error bars being the standard deviation of the averages. Figure 3. Relative (normalized) N-MIR and N-MOIR reactivities for six solvents (Kahn et al, 1997) 30

184 Figure 4. Box plots of the calculated (A) net reactivities (NRs) and (B) normalized reactivities (RAFs) across cities (Russell et al., 1995) NR * * * * * * * * RAF * * * M85 LPG Phase 2 M85 CNG LPG E85 Phase 2 CNG E85 RFA FUEL 31

185 Figure 5. Comparison of three-dimensional and trajectory modeled relative reactivities CIT Peak Ozone CIT Population Threshold Exposure CIT Spatial Threshold Exposure Box Model MIR Box Model MOIR carbon monoxide ethane benzene methyl-t-butylether 2,2,4 trimethylpentane butane methanol methyl ethyl ketone 2-methylpentane ethylbenzene toluene ethanol ethyl-t-butylether methylcyclopentane 2-methyl-1-butene o-xylene 2-methyl-2-butene ethene 3-methylcyclopentene m,p-xylene acetaldehyde 1,2,4 trimethylbenzene propene isoprene propionaldehyde+higher 1,3 butadiene formaldehyde Relative Reactivity 32

186 Figure 6. Spatial distribution of relative reactivities (Yang et al., 1999) a. Normalized HCHO Relative Reactivity (ppmo3/ppmc) at 1400 hour, August 29, 1987 in the Los Angeles, CA area. b. Coefficient of variation (relative uncertainty) (%) of the HCHO Relative Reactivity at 1400 hour, August 29,