Mathematical modeling of the concentrations of volatile organic

Aug 1, 1993 - Robert A. Harley, Armistead G. Russell, Glen R. Cass. Environ. ... Michael J. Kleeman, Lara S. Hughes, Jonathan O. Allen, and Glen R. Ca...
0 downloads 0 Views 2MB Size
Download PDF
Envlron. Sci. Technol. 1993, 27, 1638-1649

Mathematical Modeling of the Concentrations of Volatile Organic Compounds: Model Performance Using a Lumped Chemical Mechanism Robert A. Harley,+ Armlstead G. Russell,* and Glen R. Cass'gt

Environmental Engineering Science Department, Californla Institute of Technology, Pasadena, California 91 125, and Mechanical Englneerlng Department, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213 An Eulerian photochemical model is used to predict the transport and reactions of organic gas emissions to the atmosphere. The model is applied to the Los Angeles area for the August 27-29, 1987, intensive monitoring period of the Southern California Air Quality Study (SCAQS),and predictions for each of the lumped organic species in the model are evaluated using speciated organic gas concentration measurements made at nine sites. Model calculations using the official State of California emission inventory accurately predicted the observed concentrations of ethene, monoalkylbenzenes, and formaldehyde, and underpredicted the concentrations of other organic species. When the on-road vehicle hot exhaust emissions of organic gases was scaled up to approximate the actual emission rates measured in a local roadway tunnel and revisions were made to the chemical composition profiles for the organic gas emissions, model performance for both ozone and most lumped organic classes improved. 1 . Introduction

Emissions of volatile organic compounds (VOC) contribute to the formation of photochemical smog, in particular ozone. Control of VOC emissions and controls on the emissions of oxides of nitrogen (NO,) are the means by which major urban areas such as Los Angeles hope to solve their ozone air quality problems (1). The ability to account for the emissions and ambient concentrations of gas-phase organic compounds is an essential step in the development of mathematical models for use in studying regional ozone control problems. Good agreement between ozone model predictions and observations must be viewed with skepticism if predicted VOC and NO, concentrations are incorrect, yet previous attempts to model VOC concentrations generally have yielded poor results. In the few cases where model performance for VOC was examined (summarized by Tesche ( 2 , 3 ) )large , negative biases (on the order of -50 5% ) in predictions relative to observations have been seen. The source of these biases is uncertain because comparison of predictions to observations for individual organic species has not been achieved; only total VOC concentrations have been examined. The wisdom of ozone control programs that rely heavily on VOC emission reduction has been questioned in some recent studies (1, 4, 5). Significant background hydrocarbon levels may exist due to isoprene, terpenes, and other compounds emitted directly from vegetation. Because controls on anthropogenic sources of VOC cannot hope to reduce the background organics emitted from vegetation, the effect of VOC emission controls in reducing

ozone levels may not be as great as previously anticipated. In order to address definitively the issue of how important biogenic VOC emissions might be, it is important to know that emissions from both natural and anthropogenic sources can be tracked successfully. Accurate models that can predict the concentrations of specific organic compounds also are needed for other reasons. Toxic air contaminants include compounds such as benzene that are present in gasoline or are used as solvents and that are known or suspected carcinogens. As a result of recent amendments to the Clean Air Act (1990), programs to control these species will be required. The atmospheric oxidation of organic gases can produce reaction products such as formaldehyde that are likewise considered to be toxic air contaminants. The low vapor pressure reaction products of some organic gases can condense to form additional aerosol material in the atmosphere, thereby increasing particle mass concentrations and aggravating human health and visibility problems. Before accurate models for the formation of aldehydes and secondary organic aerosols can be constructed, an accurate model for the transport and chemical reactions of primary (Le., directly emitted) organics is needed. One of the reasons that detailed models for the prediction of organic gas concentrations have not been proposed and evaluated in the past is that speciated ambient VOC concentration data are scarce. During the 1987 Southern California Air Quality Study (SCAQS; for an overview, see Lawson (6)),a detailed set of speciated organic gas concentration measurements were made at a network of nine monitoring sites in the Los Angeles area. In the present study, the ability of an Eulerian photochemical air quality model to predict ambient concentrations of individual organic species and lumped species groups is evaluated. Model predictions are compared against measurements of organic species concentrations made during the August 27-29,1987, intensive monitoring period of SCAQS. The effects of present uncertainties in on-road motor vehicle hot exhaust mass emission rates are examined, and an improved description of the chemical composition of the VOC emission inventory is explored. 2. Model Description The CIT airshed model is an Eulerian photochemical air quality model that has been used to calculate the transport and chemical reactions of pollutants in the atmosphere. The theoretical basis of the CIT airshed model and its numerical implementation have been described previously (7-10). In brief, the model solves numerically the atmospheric diffusion equation for a set of reacting chemical species:

ac.+ V.(iiCi)= V.(KVCi) + Ri + Qi

2

+ California Institute of Technology. t

Carnegie Mellon University.

1636 Envlron. Sci. Technol., Vol. 27, No. 8, 1993

at where Ci is the concentration of species i, ii is the wind 0013-930X/93/0927-1638$04.00/0

0 1993 American Chemlcai Soclety

Table I. LCC. Lumped Organic Classes and Surrogate Species species code ALKA

ETHE ALKE TOLU AROM HCHO ALD2 MEK MEOH ETOH

ISOP

lumped species description

surrogate speciesb

n-butane, n-pentane, isobutane, isopentane, n-hexane, n-heptane, n-octane, 2-methylpentane, a,l-dimethylbutane, 2,3-dimethylpentane, isooctane ethene ethene propene, tram-2-butene Cs+ alkenes toluene monoalkylbenzenes di- and trialkylbenzenes m-xylene, 1,3,5-trimethylbenzene formaldehyde formaldehyde acetaldehyde Cz+ aldehydes methyl ethyl ketone ketones methanol methanol ethanol Cz+ alcohols isoprene biogenic alkenes

0 Based on the chemical mechanism of Lurmann et al. (II), with extensions to treat the alcohols and biogenic alkenes separately from the other organics (10). b Where more than one surrogate species is listed, the rate constant(s) and oxidation product yields for the corresponding lumped species are calculated based on a mixture of the listed surrogate species. c 50% of propane emissions also are assigned to the lumped alkane clase; the remaining propane emissions aretreatedasunreactive,asperLurmannetal. (11).d30%ofbenzene emissions also are assigned to the lumped alkane class,as per Lurmann et al. (11).

velocity, K is the eddy diffusivity tensor (here assumed to be diagonal), Ri is the rate of generation of species i by chemical reactions, and Qi is a source term for elevated point sources of species i. A system of coupled nonlinear equations is obtained when the above equation is written repeatedly for each of the chemical species tracked by the model. Initial conditions and lateral boundary conditions are set using measured pollutant concentration data. The ground-level boundary condition sets upward pollutant fluxes equal to direct emissions minus dry deposition. The concentration gradient for each species is set to zero at the top boundary, so that there is no vertical transport of pollutants through the top of the modeling region. Typical photochemical reaction mechanisms that are used within regional air quality models simplify the treatment of the volatile organiccompounds by aggregating the emissionsof hundreds of individual species into a much smaller number of lumped species classes. In the LCC mechanism (Lurmann, Carter, and Coyner (Ill), the organic species are lumped by molecule, with compounds having similar structure and reactivity grouped together. The chemistry of all molecules within a single lumped class is then represented using one or more surrogate species (e.g., toluene is used to represent the chemistry of all monoalkylbenzenes). In the present study, the condensed version of the LCC chemical mechanism (11) with extensions to treat the alcohols and biogenic alkenes separately (10) has been used to represent the atmospheric chemistry of the organic gases and the relavant inorganic species, including ozone and the oxides of nitrogen. The extended mechanism includes 106chemical reactions involving 35 inorganic and organic species. A list of the lumped organic species groups and the corresponding surrogate species is shown in Table I. The chemical mechanism includes separate reactions with the hydroxyl radical for each of the organic species

listed in Table I. In addition, the mechanism includes the reactions of ethene, lumped C3+alkenes, and isoprene with ozone, the nitrate radical, and atomic oxygen. Photolysis reactions involving the aldehydes and ketones also are represented in the chemical mechanism. Further details concerning the chemical mechanism used in the present study can be found elsewhere (10, 11). In this study, the LCC chemical mechanism has been further extended to track the direct emissions of formaldehyde separately from the formaldehyde that is produced by the atmospheric oxidation of other organic gases. With this modification, the model can be used to study the contributions of direct emissionsversus photochemical production to total formaldehyde concentrations. The extension of the mechanism was accomplished by introducing a specially tagged species that is used to track directly emitted formaldehyde (this tagged species participates in chemical reactions identical to those of the preexisting formaldehyde species in the LCC mechanism). Total predicted formaldehyde concentrations for comparison with ambient data are computed by summing the concentrations of the directly emitted and photochemically generated formaldehyde. The removal of pollutants at the earth's surface by dry deposition is included in the model. Dry deposition velocities are computed using local meterological, surface roughness, and land use data in each grid square (12). First, resistances ra and rb to dry deposition are calculated based solely on fluid mechanical considerations. These resistances are due respectively to turbulent transport in the atmospheric boundary layer and to molecular diffusion through a laminar sublayer near the gound. Then, a surface resistance term ( r J )specific to the pollutant and land-use type is included to account for pollutant-surface interactions: i*g

1 rb

- ra + + r:

(2)

where ugi is the actual deposition velocity for species i used in the model. Surface resistance values are derived from the recommendations of Sheih et al. (13) and engineeringjudgement (12). Land-use characteristics can be specified using detailed maps or satellite data. Dry deposition velocities for the hydrocarbons, though rarely measured, are generally thought to be low because most hydrocarbons are neither highly reactive nor highly soluble in water and therefore surface resistance terms for these species are large (14). In the present study, surface resistance values of 50 s/cm have been used for all of the lumped hydrocarbon species in the model, thereby limiting dry deposition velocities for these species to a maximum of 0.02 cm/s. Dry deposition surface resistance terms for formaldehyde and higher aldehydes are scaled relative to the surface resistance values for SOa: formaldehydesurface resistance values are set to 0.5 times the values used for SOZ; a scaling factor of 2.0 relative to SO2 is used for acetaldehyde and higher aldehydes (12). The surface resistance for PAN has been set to a value of 4.5 s/cm over land (15). The dry deposition surface resistance term for ketones has been set to a high value of 50 s/cm, the same as is used for the hydrocarbons, based on the low atmospheric reactivity of ketones. Envlron. Scl.Technol., Vol. 27, No. 8, 1903 1039

.........,.:+ ................. i....................................

r'

3800

Table 11. Comparison of Two Independent Lumped Ambient Organics Data Sets.

I...

.........

3740

,'

,... ,....................:

3700 3680

240

280

320

360

400

440

UTM E a m g (km)

480

520

560

600

Figure 1. Map of the Los Angeles,CA, area showingthe computatlonal region and the locations of special SCAQS air quality monitorlngsites.

3. Ambient Organics Concentrations

In this section, the speciated gas-phase organics concentration data acquired during SCAQS are described. These data, when lumped to match the LCC mechanism species groupings, provide the basis for comparison of organic species model predictions with observations. There are two independent data sets available. The first of these two data sets consists of six l-h average samples per day taken at the Long Beach and Claremont sites, and three l-h average samples per day at the seven additional sites shown in Figure 1, giving a total of 33 samples per day. Hydrocarbon samples were collected in evacuated stainless steel canisters and stored for later analysis (16). The samples were analyzed by gas chromatography and gas chromatography/mass spectrometry by Stockburger et al. (17)and Rasmussen (18).Carbonyl samplingwas performed using 2,4-dinitrophenylhydrazine(DNPH) impregnated cartridges (16,19). As described by Lurmann and Main (20),the above measurements were consolidated into a single data base, and the identification of the hydrocarbons was extended using relative retention time information derived by Fujita from analyses by Lonneman et al. (21). In the present study, these data will be referred to as data set 1. A second SCAQS data set consisting of speciated CZC ~ hydrocarbon O concentration measurements was reported by Lonneman et al. (21).Data set 2 consists of 3-h average samples, with measurements made at Central Los Angeles, Long Beach, and Claremont. Early morning samples were collected at all three sites at 0500 hours, with additional samples taken at the Claremont site only at 1100 hours and 1400 hours PST, for a combined total of five samples per day. A comparison of the two independent data sets via linear regression analysis is presented in Table I1 for hydrocarbons and for formaldehyde. As shown in Table 11, correlation coefficients of 0.87 or higher were computed for each of the lumped hydrocarbon species being compared. This indicates reasonable agreement between the two hydrocarbon data sets, given the different averaging times over which the samples were collected. The results shown for formaldehyde in Table I1 require further discussion. In addition to the DNPH cartridge measurements of carbonyl concentrations (19)that form part of data set 1, spectroscopic measurements of ambient formaldehyde concentrations were made at Claremont and Long Beach. Winer et al. (22)made measurements using differential optical absorption spectroscopy (DOAS), and Mackay et al. (23)used tunable diode laser absorption spectroscopy (TDLAS). The spectroscopic methods have been com1640 Envlron. Sci. Technol., Vol. 27, No. 8, 1993

species code

no. of cases

slope

intercept (ppbv)

R

ALKA ETHE ALKE TOLU AROM HCHOb LongBeach Claremont

38 34 34 37 38

0.96 f 0.09 0.96 i 0.07 1.08 f 0.07 1.26 f 0.12 0.92f 0.06

5.1 f 21 2.9 f 2.9 1.5 f 2.7 -1.8 f 2.1 -0.8 i 2.3

0.87 0.93 0.93 0.88 0.94

11 0.98 i 0.23 -0.5 f 2.1 0.81 19 0.43 f 0.25 5.8f 3.5 0.37 aFor the ALKA, ETHE, ALKE, TOLU, and AROM lumped species, the combined dataof Stockburger et al. (17) and Rasmussen (18) (data set 1) are regressed on the data of Lonneman et al. (21). All valid paired summertime samples are used in these comparisons. HCHO measurements of Fung (19) from data set 1 are regressed on the spectroscopic measurements of Winer et al. (22).

pared (221,and a correlation coefficient of 0.76 was computed for measurements made at Claremont by the two independent spectroscopic methods. Comparisons between the DOAS and DNPH-based formaldehyde measurements from the summertime monitoring periods at Claremont and Long Beach were made as part of the present study. The summertime formaldehyde measurements made at Long Beach using DOAS and DNPHimpregnated cartridges agree, whereas the measurements made at Claremont using the same two techniques are not as well correlated, as shown in Table 11. The formaldehyde data at Claremont appear in the time series plots later in this paper. Visual examination shows that the DNPH and DOAS data sets generally overlap and have similar diurnal variation, indicating that the regression results and low correlation coefficients at Claremont may be due in part to the small number of paired samples available for statistical analysis (there were 11 paired samples at Long Beach and 19 paired samples at Claremont). Ambient peroxyacetyl nitrate (PAN) concentrations were measured by Williams and Grosjean (24)using electron capture gas chromatography. Data were collected at Anaheim, Azusa, Burbank, Central Los Angeles, Claremont, Long Beach, and Rubidoux during the August intensive monitoring period of SCAQS. Independent measurements of PAN at Claremont were reported by Lonneman et al. (21).Both of these data sets appear on the time series plots later in the text. Large systematic differences between the two data sets were observed. It is likely that difficulties in calibrating the instruments are the cause of these differences, since high precision was achieved for measurements made using a single instrument (25). Therefore, large uncertainties must be associated with the ambient concentration data for PAN. The lumping of the ambient data proceeds by analogy with the method used to lump VOC emissions to the model: each individual organic species is assigned to a lumped class according to its chemical properties as per Lurmann et al. (II),and the lumped speciesconcentrations are accumulated accordingto the measured concentrations of individual species. Ambient concentrations of hydrocarbons reported in parts per billion of carbon (ppb C) were first converted to parts per billion by volume (ppbv) by dividing by the number of carbon atoms for each individual species. Then, the converted concentrations were added to the appropriate lumped molecule species group. Therefore, the comparisons between model predictions and ambient data presented later are based on

~~

~

Table 111. Upwind. Boundary Condition Values (ppb) species

co

NOz NO HCHO ALD2 MEK NMHCb

Os

boundary condition 200 1 1 3 5 4 100 40

0 Downwind boundary conditions are based on the advection flux out of the air basin. The boundary conditionat the top of the modeling region (at a height of 1100 m above ground level in this case) is a zero flux boundary condition such that pollutants are not fed into the model through the top boundary. b Non-methane hydrocarbon (NMHC) concentrationsare specifiedin ppb C instead of ppbv. Each ppb C of NMHC is speciated as follows: 0.095 ppbv ALKA, 0.017 ppbv ETHE, 0.018 ppbv ALKE, 0.015 ppbv TOLU, and 0.016 ppbv AROM.

molecular counts. This is the appropriate comparison because the LCC chemical mechanism tracks the number of emitted molecules, but does not necessarily preserve the number of carbon atoms. 4. Model Application The present study focuses on the August 27-29,1987, SCAQS intensive monitoring period. The CIT airshed model has been used to predict ozone and precursor concentrations in the Los Angeles area over the entire 3-day period (10). To perform these calculations, meteorologicaland pollutant emissionsdata are required. These data are specified for each 5 km X 5 km grid square of a regular grid system within the modeling region shown in Figure 1. The vertical extent of the modeling region is 1100 m, subdivided into five layers of thicknesses 38,116, 154, 363, and 429 m. The vertical extent chosen for the modeling region exceeds all actual measured mixing depths within the South Coast Air Basin during the 3-day period examined here. Meterological fields were developed from an extensive data base of routine and special SCAQS surface level and upper air measurements (6, 26). Hourly average wind speed and direction data were reported at 50 different stations located within the modeling region during the August 27-29 period. Upper air soundings were performed six times per day at eight special coastal and inland sites. These soundings provide a detailed characterization of vertical temperature and wind profiles in the atmosphere and were used to generate mixing depth and threedimensional wind fields. Hourly temperature measurements were reported at 59 sites, with relative humidity or dew point data also reported a t 43 of these sites. Solar ultraviolet radiometers were operated at five sites during the summer phase of SCAQS and provide some data with which to scale photolysis rate constants in the model to match observed levels of solar ultraviolet radiation (9,10, 27). This extensive set of meteorological measurements made during SCAQS provides the basis for the representation of atmospheric conditions in the model. The wind, mixing depth, temperature, humidity, and solar radiation fields were prepared by spatial interpolation of the available observations, as described in more detail elsewhere (9, 10,28, 29). The boundary condition values used in the present study are shown in Table 111. These values are based on

measurements made during SCAQS at San Nicolas Island (16) and measurements made by aircraft over the ocean upwind of the modeling region (30). An emission inventory supplied by the California Air Resources Board (ARB) was used in this study (31). Mobile source emission estimates were developed by the ARB using results from a travel demand model and the EMFAC 7E emissionsfactor model (32). Stationary source emissions were prepared by the South Coast Air Quality Management District (SCAQMD). Day-specific biogenic emission inventories were provided by the SCAQMD for August 27 and 28 based on a new gridded inventory of leaf biomass in the Los Angeles area (33). The August 28 biogenic inventory was used for both August 28 and 29 because no day-specific inventory was available for the 29th. The official inventory includes a set of 225 chemical composition profiles used to speciate the total VOC mass emissions from over 800 source types. Each speciation profile gives the detailed chemical composition in terms of weight percent for each organic species found in the total VOC emissions from a specific source type. Key speciation profiles have been reviewed and updated by Harley et al. (34). Summaries of the official and respeciated emission inventory for August 27 are presented in Table IV. The major differences between the official and respeciated VOC emission inventories are described below, with further details available elsewhere (34). The largest differences between these two VOC emission inventories are seen at the individual compound level, where for some species such as 1,3-butadiene and cyclohexane, order of magnitude changes were found in the basin-wideemissions estimates. The composition of engine exhaust VOC emissions from catalyst-equipped vehicles used in the official emission inventory was derived from the EPA 46-car study (35). The respeciated emission inventory uses results from the more recent EPA (36,37) and Auto/Oil(38) studies, and resolves compounds that co-eluted in earlier gas chromatographic analyses (e.g., benzene and cyclohexane were not resolved separately in the 46-car study). The revised speciation profile indicates higher emissions of benzene, propene, and 1,3-butadiene and lower emissionsof propane when compared to the official speciation profile. A new speciation profile for exhaust emissions from noncatalyst vehicles was developed from the results of a study of 1970-1979 model year light-duty vehicles that were tested using a current (1988) regular leaded gasoline (39). The new speciation profile includes 125 species, providing a much higher level of chemical detail than the official speciation profile which lists only 24 species. The revised speciation profile indicates higher emissions of l,&butadiene and dialkylbenzenes such as xylene and ethyltoluene. The composition of whole (liquid) gasoline and gasoline headspace vapors was measured for composite samples of gasoline sold in Los Angeles in 1984(40). Separate analyses were performed for regular and premium-grade gasoline, subdivided further into leaded and unleaded grades, and summertime and wintertime fuels. In the officialemission inventory, evaporative emissions of whole gasoline and gasoline headspace vapors are speciated using composite profiles that were calculated using 1979 gasoline volume sales by grade. Between 1979 and 1987 there was a significant increase in use of unleaded gasoline relative to leaded gasoline. Therefore, in the respeciated emission Envlron. Sci. Technol., Vol. 27, No. 8, 1993

1641

Table IV. Region-Wide. Chemical Emissions Summary for August 27, 1987 species code ALKA ETHE ALKE TOLU AROM HCHO ALD2 MEK MEOH ETOH ISOP ROG NONR

lumped species description

base case inventory (lo3kg/day) officialb revisedc speciation speciation

3X hot exhaust inventory (los kg/day) officialb revisedc speciation speciation

CI+ alkanes ethene Ca+ alkenes monoalkylbenzenes di- and trialkylbenzenes formaldehyde Cz+ aldehydes ketones methanol Cz+ alcohols biogenic alkenes reactive organic gases nonreactived carbon monoxide oxides of nitrogen

799 698 1016 884 83 60 151 114 100 104 182 190 162 118 228 178 118 146 184 218 20 19 35 31 13 13 23 24 28 49 29 51 5 1 5 1 91 223 91 235 117 117 117 117 1536 1548 2062 2042 1028 989 1175 1129 co 5624 5624 10867 10867 1138 1138 NO, 1138 1138 The inventory region includes most of the mapped area shown in Figure 1, which extends beyond the South Coast Air Basin into Ventura County and the Southeast Desert Air Basin. Using the chemical composition profiles supplied with the official emission inventory. c Using the respeciated VOC emission inventory, as per Harley et al. (34). Nonreactive compounds in the LCC mechanism include methane, ethane, acetylene, and various chlorinated compounds.

inventory, the composite whole gasoline and headspace vapor speciation profiles have been recomputed using 1987 gasoline sales data (34). This recalculation indicates 3.5% higher aromatic content in the composite liquid fuel and lower alkane content, when compared to the profile used in the official inventory. The aromatic content of gasoline headspace vapors increases slightly, by about 0.5 % ,in the revised speciation profile, but remains dominated by emissions of alkanes such as butane and pentane. Running loss evaporative emissions from motor vehicles were mislabeled as crankcase emissions in the official emission inventory and assigned to the noncatalyst vehicle exhaust speciation profile. In the revised inventory, these emissionsare speciated using the gasolineheadspace vapor profile described earlier. This results in increased emissions of C & jalkanes and reduced emissions of larger alkanes, aromatics, and combustion-derived species such as ethene and acetylene. Revised speciation profiles also have been developed for solvent emissions associated with industrial surface coatings, industrial adhesives, water-borne and organic solvent-borne architectural surface coatings, and thinning solvents (34). These revised speciation profiles are based on the use of solvents reported by the surface coating manufacturers, in place of analyses of finished surface coating products which form the basis of the speciation profiles in the official emission inventory. The revised speciation profiles indicate increased emissions of petroleum-derived CB-Clz alkanes and cycloalkanes and increased emissions of alcohols and glycols. The aromatic fraction of solvent emissions from surface coatings is lower in the respeciated emission inventory when compared to the official inventory. In the official emissioninventory, in cases where detailed knowledge of the chemical composition of the VOC emissions from a particular source was not available, the emissions were speciated using a ‘speciesunknown’ profile that is a composite of the basin-wide emissions from all sources. In the respeciated inventory (341,the total mass of VOC emissions assigned to this profile has been reduced by 48% by reassigning the emissions from some sources to other more appropriate speciation profiles. The sources 1842 Envlron. Scl. Technol., Vol. 27, No. 8, 1993

that have been reassigned include organic solvent emissions from surface coating thinning and cleanup (16 tonfday of VOC emissions) and from cleaning and pretreatment of vehicles prior to repainting (38ton/day of VOC emissions). There are currently no appropriate alternative speciation profiles for the remaining ‘species unknown’ VOC emissions from sources such as commercial cooking (charbroiling, deep fat frying, and other unspecified emissions), manufacture of plastics and plastic products, and a large number of unusual sources with less than 1 tonfday of emissions. VOC emissions from these sources may include a range of complex organic molecules that were not included in the ambient concentration measurements and probably do not resemble the speciation of the composite inventory for the entire air basin which is dominated by mobile source and surface coating emissions and light hydrocarbons such as methane. Therefore, in the respeciated VOC emission inventory, the remaining ‘species unknown’ emissions have been assigned to the lumped Cz+ alcohol species to prevent interferences with the analysis of model performance for the lumped hydrocarbons, while still providing an approximate representation of the reactivity of these emissions. Development of appropriate speciation profiles for these sources would provide the best solution to uncertainties in the chemical composition of these VOC emissions. About half of the increase in Cz+ alcohol emissions shown in Table IV can be attributed to the assignment of ‘species unknown’ emissions to this lumped class. The use of revised speciation profiles for VOC emissions from industrial surface coatings and water-borne architectural surface coatings also contributes to the increase in CZ+alcohol emissions beyond the amount suggested in the official emission inventory. Both the official and respeciated emission inventories contain a separate accounting of over 250 organic compounds, which provides a much higher level of chemical detail than is carried in the lumped mechanism used in these photochemical modeling calculations. Sometimes there are a small number of species which dominate the emissions assigned to a lumped organic class: for example, the lumped monoalkylbenzene class is dominated by

toluene (approximately 90 74 of assigned emissions), with smaller amounts of ethylbenzeneand propylbenzene. The di- and trialkylbenzenes are dominated by emissions of isomers of xylene, ethyltoluene, and trimethylbenzene. The ETHE species used in the model represents ethene emissions almost exclusively, with small amounts of chlorinated alkenes also included (these chlorinated alkenes account for less than 1%of assigned emissions). The lumped alkene class is dominated by emissions of propene, with smaller amounts of l-butene and other isomers of butene and pentene. In contrast, the lumped alkane class encompasses significant emissions of many different straight-chain and branched alkanes as well as cycloalkanes. The aldehyde and ketone species classes are used in the LCC mechanism to represent both the direct emissions of such compounds as well as their formation in the atmosphere as the oxidation products of other emitted organic species. The aldehyde emissions are made up mostly of formaldehyde and acetaldehyde, with smaller amounts of other aldehydes including propionaldehyde,benzaldehyde, and acrolein (propenal). The ketone emissions are made up of acetone, methyl ethyl ketone, and methyl isobutyl ketone. Only small amounts of methanol are included in the current VOC emission inventory. Emissions of higher alcoholsinclude significant amounts of ethanol, 2-propanol, and glycols and glycol ethers from stationary sources such as surface coating and solvent use. The emissionsof biogenic alkenes, which are lumped separately from other alkenes in the model, are dominated by isoprene, a-pinene, and P-pinene. In addition to using both the official and respeciated VOC emission inventories in this study, a sensitivity analysis of the effects of using increased on-road vehicle hot exhaust emissions has been performed. Based on measurements made in a roadway tunnel during SCAQS (41),it was found that actual emissions of carbon monoxide and organic gases from motor vehicles in Los Angeles were significantly higher than suggested by the EMFAC model (41,42).To account for a possible understatement of such vehicular emissions, alternate emission inventories were created in which the carbon monoxide and organic gas hot exhaust emissions from on-road vehicles were increased to three times the base case values, as suggested by the results of the tunnel study. This change represents an increase in reactive organic gas emissions of about 500 tonsiday over the entire mapped region shown in Figure 1. In order to apply the LCC chemical mechanism to a particular urban area, there are three parameters that should be set to match local conditions. The parameters are as follows: the fraction of total C4+alkanes made up of CrC6 alkanes, the terminal alkene fraction of total C3+ alkenes, and the dialkylbenzene fraction of total di- and trialkylbenzenes. Default values for these parameters specified by the LCC mechanism are 0.43,0.60,and 0.60, respectively, but Lurmann et al. (II) recommended that these parameters should be set using location-specific emissions data or speciated ambient concentration measurements. These parameters affect both the reaction rate constants and the oxidation product yields for the corresponding lumped hydrocarbon species. In the present study, these parameters were calculated using the distribution of species indicated in the organic gas emission inventory for the South Coast Air Basin, as described below.

The default value (0.43)for the C . & fraction i of total C4+alkanes was used in all cases and was not significantly different from the values indicated by analysis of the official and respeciated emission inventories. The terminal alkene fraction of total C3+ alkenes is ambiguous in the official emission inventory because a large fraction of alkene emissionsare listed as unspecified isomers of butene and pentene, without distinction between isomers with terminal and internal double bonds. The respeciated VOC emission inventory is more specific in listing exactly which alkene isomers are emitted, and although some ambiguity remains, the default value appears to agree with the value calculated using the respeciated emission inventory. Therefore, the default terminal alkene fraction (0.60)of total C3+ alkenes recommended by Lurmann et al. (11) was used in the present study. In the case of the di- and trialkyl benzenes, the default dialkylbenzenefraction (0.60) was not used in the present study. Values of this parameter were calculated to be 0.72 using the official emission inventory and 0.81 using the respeciated emission inventory. This implies that the dialkylbenzenes comprise a larger fraction of total di- and trialkylbenzene emissions in the Los Angeles area than is reflected by the default value of the correspondingparameter in the LCC chemical mechanism. 5. Results

The CIT airshed model has been applied to the August 27-29 period using the meteorological and emissions input data described above. In this section, statistical and graphical comparisons of model predictions and observations are presented. Statistical model performance measures are presented in Table V for ozone, total reactive hydrocarbons (RHC), and NO2 concentrations (nitrogen dioxide plus other nitrogen-containing species that are measured as if they are NO2 by chemiluminescent NO, monitors). These statistical evaluations are based on data for August 28 only because August 29 was a Saturday, and the emission inventory for southern California does not contain a separate accounting of weekend traffic patterns and emissions, as was explained earlier. A more detailed analysis of prior model performance for these species, including time series plots of observed and predicted concentrations, is available elsewhere (IO). Results given in Table V show an improvement in model performance of 1-2% relative to the results in ref 10 due to changes made to the vertical advection calculations in the model. Perturbation analysis employing changed boundary and initial conditions likewise is reported elsewhere (9) and shows that model results for August 27 are sensitive to changes in initial conditions but that predictions for August 28 and 29 are not. Perturbation of the model using very clean (i.e., nearly zero) boundary conditions likewise does not affect predicted urban pollutant concentrations significantly, as the base case boundary conditions originally are quite low (9). In the base case calculation using the official emission inventory, ozone and RHC concentrations are underpredicted (normalizedbiases of -23 % and -35 % ,respectively), whereas NO2 concentrations are overpredicted by 1874 on average. Predicted ozone and NO2 concentrations do not Environ. Scl. Technol., Voi. 27, No. 8, 1993 1648

Table V. Model Performance for Ozone, NOz, and Total Reactive Hydrocarbons on August 28

statistical measuree biad normalized bias gross errorf gross error (normalized)

species ozone NO2 RHC ozone NO2 RHC ozone NOz RHC ozone NO2 RHC

base case inventow officialc revisedd speciation speciation -3.1 +0.2 -56 -23 %

+18% -35 % 4.1 1.7 71 36 % 41 % 51%

-3.2 +0.1 -69 -24 % +17% -46 % 4.2 1.7 80 36 % 41 % 57 %

3X hot exhaust inventory* official' revisedd speciation speciation -0.4 +0.4 -29 +1% +23 % -12% 2.9 1.9 59 27 % 45 % 47 %

-0.7 +0.4 -46 -2 % +22% -26 % 2.9 1.8 70 27 % 44 % 53 %

a From the CIT airshed model using the base case total mass emissions. From the CIT airshed model using scaled up on-road vehicle hot exhaust emissions. Using the chemical composition profiles supplied with the official inventory. d Using the respeciated VOC emiaaion inventory of Harley et al. (34). e Bias is defined as the mean residual (predicted minus observed) concentration. Gross error is the mean of the absolute residuals. Normalized statistics are calculated by dividing each residual by the corresponding observed concentration before averaging. Cutoff concentrations of 6 pphm and 2 pphm are used for ozone and NO2 respectively. No cutoff is used for the hydrocarbons. f Unnormalized bias and gross error statistics are stated in units of parts per hundred million (pphm) for ozone and NO2 and in units of parts per billion by volume for total reactive hydrocarbons (RHC). f

change significantly when the revised speciation profiles are used in place of the official speciation profiles within the base case emission inventory, but the underprediction of RHC concentrations becomes larger, as shown in Table V. This is due to the increases in alcohol and ketone emissions and simultaneous decreases in alkane and monoalkylbenzene emissions in the respeciated inventory compared with the official base case emission inventory (see Table IV). Note that alcohol, aldehyde, and ketone species are not included in the analysis of model performance for total RHC because these species are not, strictly speaking, hydrocarbons and because alcohol species concentrations were not measured during SCAQS. Table VI shows statistical comparisons of model performance for each of the lumped species defined in Table I except the alcohols for which no ambient data are available and the biogenic alkenes, because the monoterpenes were not measured in the ambient data sets (only isoprene was detected). All performance statistics shown in Table VI were calculated using observed organic species concentrations from data set 1. Data set 1provides a large number of samples and good spatial coverage of the modelingregion for the SCAQSintensive monitoring days, whereas only five samples are available from data set 2 on any given day. Observations from data set 2 are shown on the time series plots presented later in the text, and these data are generally consistent with observations from data set 1. For the base case simulation using the official State of California emission inventory and emission speciation profiles, the statistical performance analysis shows that the biases in model predictions for the lumped organic species tend toward underprediction in all cases. The underpredictions are largest (about -50 % normalized bias) for the di- and trialkylbenzenes, C3+ alkenes, and CZ+ aldehydes; for all remaining species other than PAN, the underpredictions are less (-1% to -26 % normalized bias). At this stage of model development for organics, examination of such bias statistics is particularly important because the key modeling issues focus on the overall bias in organic compound emission inventories (e.g., are the overall estimates of the emissions of particular organic compound classes too high or too low?). Gross error 1644 Envlron. Scl. Technol., Vol. 27, No. 8, 1993

statistics are also given in Table VI that exclude the effect of averaging overpredictions and underpredictions and that, therefore, capture added effects such as errors in the timing of emissions into the model. When the respeciated VOC emission inventory (34) is used in place of the official inventory, predicted alkane, ethene, and monoalkylbenzene concentrations decrease, while di- and trialkylbenzene concentrations increase, and the remaining lumped species stay about the same. Predicted ozone concentrations decreasevery slightly when the respeciated VOC emission inventory is used, as shown in Table V. When the calculations are repeated using the official speciation profiles and increased on-road vehicle hot exhaust emissionsas suggested by the SCAQStunnel study (41),predicted ozone and RHC concentrations increase to match observed values more closely as shown in Table V. Predicted NO2 concentrations increase slightly. Detailed analysis of the various lumped organic compounds shown in Table VI indicates that model performance generally improves, except for ethene and formaldehyde which become overpredicted. Finally, when both the increased exhaust emissions and the revised emissions composition profiles are used simultaneously, the overprediction of ethene and formaldehyde concentrations is reduced. Concentrations of all other lumped organic species are predicted with a normalized bias of i30% or less in this case. The absolute gross errors (in ppbv) for half of the lumped organics examined improve relative to the base case, while the other half do not. Model performance clearly depends critically on both VOC composition and total mass emission rates. There is little change in predicted ozone and NO2 concentrations compared to the calculations using increased exhaust emissions and the official emission speciation profiles. The increments in predicted RHC concentrations that occur when the revised speciation profiles are substituted for the official emission speciation profiles in the presence of the increased motor vehicle hot exhaust emissions are similar to those found previously when the base case emission inventory was perturbed in the same manner and, again, are due to the changes in the

Table VI. Model Performance for Lumped Organic Species on August 28 base case inventov species officialc revisedd statistical measuree code speciation speciation

3X hot exhaust inventoryb offici3 revisedd Speciation speciation

-30 -18 -38 -28 ALKA +2.7 +6.6 -4.7 -2.7 ETHE -1.3 -1.1 -4.1 -4.2 ALKE -2.0 -0.2 -4.5 -3.0 TOLU -2.9 -3.8 -4.5 -5.3 AROM +1.9 -2.0 +3.0 -1.7 HCHO -6.9 -6.3 -10.0 -9.9 ALD2 -4.6 -4.7 -5.9 -6.3 MEK -1.0 -3.1 -0.9 -3.2 PANf -29 % -12 % -41 % -26 % ALKA normalized bias +41% +74% -26 % -8 % ETHE -7 % -5 % -45 % -46 % ALKE +2% +22% -29 % TOLU -13 % -22 % -35% -45 % -56 % AROM +36 % +46 % -3 % -1 % HCHO -21% -17% -46 % ALDB -45 % -3 % -5 % -10% -14% MEK +lo% +8 % -33 % -34 % PANf 44 36 41 48 ALKA gross error (ppbv) 5.3 7.0 4.9 6.4 ETHE 5.1 5.5 5.1 5.5 ALKE 4.4 3.9 5.2 4.0 TOLU 4.4 5.1 5.1 5.8 AROM 3.6 3.5 4.1 3.3 HCHO 8.3 10.3 8.1 10.2 ALDB 6.8 7.6 6.5 7.5 MEK 2.1 3.4 2.1 3.5 PANf gross error (normalized) ALKA 51 % 55% 45 % 47 % 79 % 60 % 53 % ETHE 43 % ALKE 71 % 71 % 78 % 78 % TOLU 38 % 46 % 51% 49 % 64 % AROM 67 % 56 % 54 % HCHO 32 % 35 % 53 % 47 % ALD2 54 % 54 % 49 % 49 % MEK 41 % 43 % 39 % 42 % PAN/ 46 % 46 % 38 % 38 % From the CIT airshed model using the base case total mass emissions. From the CIT airshed model using scaled up on-road vehicle hot exhaust emissions. c Using the chemicalcompositionprofiles supplied with the officialinventory. Using the respeciated VOC emissioninventory of Harley et al. (34). e Bias is defined as the mean residual (predicted minus observed) concentration. Gross error is the mean of the absolute residuals. Normalized statistics are calculated by dividing each residual by the corresponding observed concentration before averaging. Only observations from data set 1 are used when computing these statistics. f A cutoff concentration of 2 ppb was used when computing performance statistics for PAN to avoid overemphasis of errors when observed concentrationsare low (Le., at night). No cutoff is used for the other species because lumped concentration measurements were typically well above detection limits and because the number of available data points is limited.

bias (ppbv)

emissions of oxygenated organics that are not included in the RHC data. Time series plots of observed and predicted concentrations for the entire 3-day period are shown in Figure 2 for the alkanes, in Figure 3 for ethene, and in Figure 4 for di- and trialkylbenzenes. The model predictions plotted employ the revised speciation profiles. The monitoring sites shown in these figures represent a transect drawn from west to east across the air basin from Hawthorne at the coast through central Los Angeles to Claremont and Rubidoux inland. Similar trends in the comparisons between model predictions and observations are seen at the other four sites where speciated organics measurements were made (although measurements also were made at San Nicolas Island, this site lies outside of the modeling region and so comparisons with observations at this site were not possible). Nighttime model predictions at Long Beach are higher than the observed concentrations, because stable conditions and highly restricted vertical mixing were assumed by the model at night, while actual temperature soundings a t night in this location show that the atmosphere was neutrally stratified, possibly due to urban heat island effects (IO).

Predicted and observed concentrations of the lumped alkane species (see Figure 2) are the highest of all lumped organic species defined in the LCC mechanism. Concentrations tend to be underpredicted in all of the cases considered in this study, and the changes in model predictions are relatively small for the lumped alkane species when on-road vehicle exhaust emissions are increased. Ethene concentrations shown in Figure 3 show good agreement between the base case predictions and observations; overpredictions are seen for the case with increased motor vehicle exhaust emissions. Both model predictions and observations show that peak ethene concentrations occur during the morning rush hour. The di- and trialkylbenzenes plotted in Figure 4 also exhibit peaks in predicted and observed concentrations during the morning rush hour. The concentrations of this lumped aromatic species tend to be underpredicted in the base case calculation; the underpredictions are reduced when on-road vehicle exhaust emissions are scaled up. Both measured and predicted concentrations of di- and trialkylbenzenes fall to their lowest values during the early to mid-afternoon hours. Envlron. Scl. Technol., Vol. 27, No. 8, 1993 1045

"""

250

d

c

1

Hawhorne

I

loot

300

250 .

I

I

I

Bo 50

f

-

1

Hawlhom

40*

I

1

Central LA

g 200 -

-

3 150-

.....

......,. 50 .

f ' I

250 .

40-

0,

.."'. ........ ...........G . 300

Claremont

0 .

c ,

-

Rubidoux

P

August 27

-

Tim (PST) August 28

August 29

~~

......

I

Rubidoux

Base Case Model 3x Hot Exhaust Model

-

Obsewed (Data Set #I) Obsewed (Data Set w2)

Flgure 2. Time series plots of observed C4+ alkane concentrations (measurements from data set 1 plotted as solid circles; 3-h average samples fromdata set 2 plottedas horlrontalbars)andmodel predlctions for the base case (solld line) and for the case of increased on-road vehicle hot exhaust emisslons (dashed line). Only results obtalned using the respeciated VOC emission Inventory (34are shown.

Base case model predictions and observed formaldehyde concentrations are shown in Figure 5. The observed and predicted formaldehyde concentrations were higher at the inland sites. Only model results using the updated VOC emissions speciation profiles and the base case mass emissions are shown in this figure. Minor differences in model results for formaldehyde were found when the official emission inventory speciation profiles were used. The directly emitted component of total formaldehyde concentrations is also presented in Figure 5, based on model calculations using the extended chemical mechanism described previously. Inspection of Figure 5 shows that most of the atmospheric formaldehyde during this summertime episode is formed by atmospheric reactions rather than being directly emitted. Similar behavior is seen throughout the land area included in the modeling region. The percentage contribution of direct emissions to total formaldehyde concentrations was found to increase overnight, and to reach a peak during the morning rush hour. The contribution of direct formaldehyde emissions becomes negligible during the late morning hours through late afternoon. The relative importance of direct formaldehyde emissions is predicted to be highest a t the coastal (upwind) sites. 1646 Envlron. Scl. Technol., Vol. 27, No. 8, lB93

4 4

August 27

......

Time (PST) August 28

Base Case Model 3X Hot Exhaust Model

*

-

August 29 Obsewed (Data Set #I) Obsewed (Data Set #2)

Flgure 9. Time series plots of observedethene concentrations(plotted as solld circles and horizontal bars) and model predictions for the base case (solld ilne) and for the case of Increased on-road vehicle hot exhaust emissions (dashed line). Only results obtalned using the respeclated VOC emission inventory (34are shown.

Model predictions and observations for peroxyacetyl nitrate (PAN) are shown in Figure 6. The diurnal pattern and nighttime concentrations seen in the observed data are reproduced by the model. Predicted PAN concentrations agree approximately with observations at central LA and Rubidoux, but daytime peak concentrations are underpredicted at Claremont, as well as at some of the other inland sites. The model predictions shown in Figure 6 were calculated using the respeciated VOC emission inventory. Predicted PAN concentrations were not significantly different when the official emission inventory speciation profiles were used. The underpredictions of PAN concentrations at downwind monitoring sites could be due to a combination of factors including inadequate photochemical production, or understated emissions of acetaldehyde (a PAN precursor), or a general understatement of all organic gas emissions into the model. The LCC mechanism includes other PAN analogs in the predicted PAN concentrations, so if anything the underprediction of observed peak PAN concentrations is larger than the performance statistics suggest due to approximations in the LCC mechanism. In comparisons of predictions made using the LCC mechanism with results of smog chamber experiments, good agreement between measured and predicted ozone concentrations was found, but predictions of other secondary species such as PAN

401

Hawthorne

25

-

20-

-

401

I

!.,

-

0

1

A

8

20-

4

I

_..-.,,.-.......,

,.... .._... ......

...s

;

>'

*.* *

1

8

12 18 20 24

Central LA

,,.... ..**.... , e

*..__.e

I

1

1

4

8

1

1

1

1

12 16 20 24

1

1

4

8

*

1

,

1

12 16 20 24

I

Clarernont

l -.

...

.

I

I

L

00 40

30

25

30b

i 2 0 1

F

-

Central LA

30

Clarernont

25

.-

I

-"I

I

Rubidoux

I

I

Rubidoux

I .-. Time (PST)

lime (PST) August 27

......

August 28

Base Case Model 3X Hot Exhaust Model

-

August 29

August 27

August 28

Observed (Data Set 111)

-

Base Case Model

Observed (Data Set #2)

_._.-

Directly Emitted HCHO

Flgure 4. Time series plots of observed di- and trialkylbenzene concentratlons(plotted as solid circles and horizontalbars) and model predlctlons for the base case (solid line) and for the case of increased on-road vehlcle hot exhaust emissions (dashed llne). Only results obtalned usingthe respeclatedVOC emlsslon lnventory(34are shown.

and formaldehyde were less accurate (11). Therefore, a final possible explanation for differences between predicted and observed PAN concentrations in the present study is uncertainty in the LCC chemical mechanism itself.

6. Conclusions The application of a model to predict the transport and chemical reactions of volatile organic compounds in the atmosphere has been described for the August 27-29 SCAQS episode. In the model, the many individual organics are aggregated into 11 lumped organic species classes. Model predictions have been compared with speciated organic gas concentration measurements made during SCAQS that have been lumped according to the same procedures used to lump organic gas emissionswithin the model. For the base case calculation using the official State of California emission inventory and chemical speciation profiles, ethene, monoalkylbenzenes, MEK, and formaldehyde concentrations were predicted with a normalized bias of -1 5% to -14 % , while the ozone and other lumped organic species concentrations were underpredicted. The largest underpredictions for organics were found for the di- and trialkylbenzenes, C2+ aldehydes, C3+ alkenes, and PAN.

August 29 Observed (DNPH)

0

ObSeNed (DOAS)

Figure 5. Time series plots of observedformaldehyde concentrations (plotted as solid circles (DNPH) and open circles (DOAS)) and model predictions using the base case emissions (solld line). The directly emltted component of predicted formaldehyde concentratlonsIs also shown (dashdotted line). Only the base case caiculatlon using the respeclated VOC emlsslon inventory ( 3 4 is shown.

Introduction of updated VOC emission speciation profiles resulted in significant changes in model performance for the organics, while predicted ozone concentrations did not differ appreciably from those obtained using the VOC speciation profiles supplied by the government with the official emission inventory. Model results for a few lumped organic species using the base case inventory indicated larger biases and gross error values when the revised speciation profiles were used in place of the official profiles. However, since the total VOC mass emissions in the base case inventory are probably understated, it should not be concluded that the revised speciation profiles are inferior to the official speciation profiles. When the on-road vehicle hot exhaust emissions were increased to reflect emission rates measured in a roadway tunnel during SCAQS (41,421, much better agreement with observed ozone concentrations was obtained. As shown in Table V, predicted ozone concentrations were sensitive to the change in VOC mass emissions (IO), but not to the change in chemical composition of VOC emissions. Similarly, predicted NO2 concentrations were somewhat sensitive to the change in VOC mass emissions, but not to the changes in VOC emission speciation. Envlron. Scl. Technol., Vol. 27, No. 8, 19B3 1647

30

ical formation in the atmosphere is amore important source of formaldehyde than direct emissions, especially during the late morning and all afternoon hours. Therefore, any program to control formaldehyde concentrations in the Los Angeles atmosphere must consider photochemical formation pathways in addition to the control of direct emission sources.

Hawthorne

25

30

Acknowledgments The authors wish to thank Paul Allen, Bart Croes, and Eric Fujita of the California Air Resources Board for helpful discussionsand for supplying SCAQS-relateddata. The authors thank David Chock for his detailed review of the CIT airshed model source code and helpful suggestions and comments. This research was supported initially by the Coordinating Research Council under Project SCAQS8. Development of the respeciated VOC emission inventory and application of the model to make use of that information was supported by the Electric Power Research Institute under Agreement RP3189-3.

Claremont

25

8

3a

20 l5

10 5

0'

4

12 16 20 24

8

4

8

12 16 20 24

4

8

12 16 20 24

Aubidoux

i9

2o 15-

..

U

..

lime (PST) August 27

l-

........

August 28

August 29

Base Case Model 3X Hot Exhaust Model

Observed (DGA) 0

ObSeNed (EPA)

Flgure 6. Time series plots of observed peroxyacetyl nitrate (PAN) concentrations (measurementsreported by Williams and Grosjean (24 plotted as solid clrcles; measurements reported by Lonneman (21) at Claremont plotted as open clrcles) and model predictions for the base case (solid line) and for the case of Increased on-road vehicle hot exhaust emlssions (dashed line). Only results obtained using the respeclated VOC emission Inventory (34are shown. Note that no observations were made at the Hawthorne site for this period.

However, the change in both VOC mass emissions and VOC chemicalcompositiondid result in significant changes in model performance for the lumped organic species. Use of the increased motor vehicle hot exhaust emissions plus revised speciation profiles results in normalized bias values for alkanes, C3+ alkenes, monoalkylbenzenes, higher aromatics, Cz+ aldehydes, MEK, and PAN that are in the range f 3 0 % or less. Formaldehyde and ethene concentrations are overpredicted in this case with normalized biases of 36% and 41 % ,respectively. The effect of these emission changes on gross error statistics is more variable: the absolute gross error (in ppb) for half of the lumped organic species improves relative to the base case, while for the other half it does not. This first attempt at model verification for lumped organic species concentration predictions has produced encouraging results. The remaining differences between observed and predicted lumped organics concentrations may be useful in diagnosing particular problems that remain in the underlying emission inventory. Model calculationsindicate that during this summertime air pollution episode in the Los Angeles area, photochem1848

Environ. Scl. Technol., Vol. 27. No. 8, 1993

Literature Cited (1) National Research Council. Rethinking the ozone problem in urban and regional air pollution. National Academy Press: Washington, DC, 1991. (2) Tesche, T. W.; Georgopoulos, P.; Seinfeld, J. H.; Cass, G. R.; Lurmann, F. W.; Roth, P. M. Improvement of procedures for evaluating photochemical models. Radian Corp., Sacramento. CA. 1990. ReDort to the California Air Resources Board under' Contract A832-103. Tesche, T. W.Alpine Geophysics,Crested Butte, CO, 1991. Personal communication. Milford, J. B.; Russell, A. G.; McRae, G. J. Environ. Sci. Technol. 1989,23,1290-1301. Chameides, W. L.; Lindsay, R. W.; Richardson, J.; Kiang, C. S. Science 1988,241,1473-1475. Lawson, D. R. J.Air Waste Manage. Assoc. 1990,40,156165. McRae, G. J.;Goodin, W. R.;Seinfeld, J. H.Atmos.Environ. 1982,16,679-696. Russell, A. G.; McCue, K. F.; Cass, G. R. Environ. Sci. Technol. 1988,22,263-271. Harley, R. A.; Russell, A. G.; McRae, G. J.; McNair, L. A.; Winner, D. A.; Odman, M. T.; Dabdub, D.; Cass, G. R.; Seinfeld, J. H. Continued development of a photochemical model and application to the Southern CaliforniaAir Quality Study (SCAQS) intensive monitoring periods: Phase I. Cargnegie Mellon University, Pittaburgh, PA, and California Institute of Technology, Pasadena, CA, 1992. Report to the Coordinating Research Council under Project SCAQS8. Harley, R. A.; Russell, A. G.; McRae, G. J.; Cass, G. R.; Seinfeld, J. H . Enuiron. Sci. Technol. 1993,27, 378-388. Lurmann, F.W.; Carter, W. P. L.; Coyner, L. A. A surrogate species chemical reaction mechanism for urban-scale air quality simulation models. Volumes I and 11. ERT Inc., Newbury Park, CA, and Statewide Air Pollution Research Center, University of California, Riverside, CA, 1987. Report to the U.S. Environmental Protection Agency under Contract 68-02-4104. Russell, A. G.; Winner, D. A.; McCue, K. F.; Cass, G. R. Mathematical modeling and control of the dry deposition flux of nitrogen-containing air pollutants. EQL Report 29, Environmental Quality Laboratory, California Institute of Technology, Pasadena, CA, 1992. Report to the California Air Resources Board under Contract A6-188-32. Sheih, C. M.; Wesely, M. L.; Walcek, C. J. A dry deposition module for regional acid deposition. Atmospheric science8 laboratory, US. Environmental Protection Agency, Research Triangle Park, NC, 1986. EPA-600/3-86-037. Wesely, M.L. Atmos. Environ. 1989,23,1293-1304. Garland, J. A.; Penkett, S. A. Atmos. Environ. 1976,18, 1737-1750.

(16) Chan,M.;Durkee, K. Southern CaliforniaAir Quality Study B-Siteoperations. AerovironmentInc., Monrovia,CA, 1989. Report tothe CaliforniaAir ResourcesBoard under Contract A5-196-32. (17) Stockburger, L.; Knapp, K. T.; Ellestad, T. G. Overview and analysis of hydrocarbon samples during the summer Southern California Air Quality Study. Paper 89-139.1, Presented at the 82nd annual meeting of the Air and Waste Management Association,Anaheim, CA, 1989. (18) Rasmussen, R. A. SCAQS hydrocarbon collection and analyses (Part I). Biospherics Research Corp., Hillsboro, OR, 1990. Report to the California Air Resources Board under Contract A6-179-32. (19) Fung, K. Carbonyl observationsduring the SCAQS. Paper 89-152.3, Presented at the 82nd annual meeting of the Air and Waste Management Association,Anaheim, CA, 1989. (20) Lurmann, F. W.; Main, H. H. Analysis of the ambient VOC datacollected in the Southern CaliforniaAir Quality Study. SonomaTechnologyInc., Santa Rosa, CA, 1992. Report to the California Air Resources Board under Contract A832130. (21) Lonneman, W. A.; Seila, R. L.; Ellenson, W. Speciated hydrocarbon and NO, comparisons at SCAQS source and receptorsites. Paper 89-152.5,Presented at the82ndannual meeting of the Air and Waste Management Association, Anaheim, CA, 1989. (22) Winer, A. M.; Biermann, H. W.; Dinoff,T.; Parker, L.; Poe, M. P. Measurements of nitrous acid, nitrate radicals, formaldehyde, and nitrogen dioxide for SCAQSby differential optical absorption spectroscopy. Statewide Air Pollution Research Center, University of California, Riverside, CA, 1989. Report to the California Air Resources Board under Contract A6-146-32. (23) Mackay,G. I.; Karecki, D. R.; Schiff, H. I. SCAQS: tunable diode laser absorption spectrometer measurements of HzOz and H&O at the Claremont and Long Beach "A" sites. Unisearch Associates, Concord, Ontario, Canada, 1988. Report tothe CaliforniaAir ResourcesBoard under Contract A732-073. (24) Williams, E. L.; Grosjean, D. Atmos. Environ. 1990,24A, 2369-2377. (25) Fujita, E. M. Research Division, California Air Resources Board, Sacramento, CA, 1991. Personal communication. (26) Croes, B. E. Research Division, California Air Resources Board, Sacramento, CA, 1990. Personal communication. (27) Zafonte, L.; Rieger, P. L.; Holmes, J. R. Enuiron. Sci. Technol. 1977,11,483-487. (28) Goodin, W. R.; McRae, G. J.; Seinfeld, J. H. J. Appl. Meteorol. 1979, 18, 761-771. (29) Goodin, W. R.; McRae, G. J.; Seinfeld, J. H. J. Appl. Meteorol. 1980, 19, 98-108.

(30) Main, H. H.; Lurmann, F. W.; Roberts, P. T. Pollutant concentrations along the western boundary of the South Coast Air Basin. Part I Areview of existingdata. Sonoma Technology,Inc., SantaRosa, CA,1990. Report to the South Coast Air Quality Management District. (31) Wagner, K. K.; Allen, P. D. SCAQSemissions inventory for August 27-29, 1987 (Tape ARA714). Technical Support Division, CaliforniaAir Resources Board, Sacramento, CA, 1990. Personal communication. (32) Yotter, E. E.; Wade, D. L. Development of a gridded motor vehicle emission inventory for the Southern CaliforniaAir Quality Study. Paper 89-137.2, Presented at the 82nd annual meeting of the Air and Waste Management Aesociation, Anaheim, CA, 1989. (33) Horie, Y .; Sidawi,S.; Ellefsen, R. Inventory of leaf biomass and emission factors for vegetation in the South Coast Air Basin. Valley Research Inc., Van Nuys, CA, 1990. Report to the South Coast Air Quality Management District under Contract 90163. (34) Harley, R. A.; Hannigan, M. P.; Cass, G. R. Environ. Sci. Technol. 1992,26, 2395-2408. (35) Sigsby,J. E.; Tejada, S.; Ray, W.; Lang, J. M.; Duncan, J. W. Environ. Sci. Technol. 1987,21, 466-475. (36) Stump,F.;Tejada,S.;Ray,W.;Dropkin,D.;Black,F.;Crews, W.; Snow, R.; Siudak, P.; Davis, C. 0.;Baker, L.; Perry, N. Atmos. Environ. 1989,23, 307-320. (37) Stump,F.;Tejada,S.;Ray, W.;Dropkin,D.;Black,F.;Snow, R.; Crews, W.; Siudak, P.; Davis, C. 0.;Carter, P. Atmos. Enuiron. 1990,24A, 2105-2112. (38) Burns, V. R.; Benson, J. D.; Hochhauser, A. M.; Koehl, W. J.; Kreucher, W. M.; Reuter, R. M. SAE Tech. Pap. Ser. 1991, No. 912320. (39) Cohu, L. K.; Rapp, L. A.; Segal,J. S. EC-1emission control gasoline. ARC0 Producta Co., Anaheim, CA, 1989. (40) Oliver,W. R.; Peoples, S. H. Improvement of the emission inventory for reactive organic gases and oxides of nitrogen in the South Coast Air Basin. Systems Applications, Inc., San Rafael, CA, 1985. Report to the California Air Resources Board under Contract A2-076-32. (41) Ingalls, M. N.; Smith, L. R.; Kirksey, R. E. Measurement of on-road vehicle emission factors in the California South Coast Air Basin. Volume I: Regulated emissions. Southwest Research Institute,San Antonio, TX, 1989. Report to the Coordinating Research Councilunder Project SCAQS1.

(42) Pierson, W. R.; Gertler, A. W.; Bradow, R. L. J . Air Waste Manage. Assoc. 1990,40, 1495-1504. Received for review September 11, 1992.Revised manuscript received March 22, 1993. Accepted March 23, 1993.

Envlron. Scl. Technol., Vol. 27, No. 8, 1993 1640