Particulate matter speciation profiles for light-duty gasoline vehicles in the United States

Abstract

Representative profiles for particulate matter particles less than or equal to 2.5 µm (PM_2.5) are developed from the Kansas City Light-Duty Vehicle Emissions Study for use in the U.S. Environmental Protection Agency (EPA) vehicle emission model, the Motor Vehicle Emission Simulator (MOVES), and for inclusion in the EPA SPECIATE database for speciation profiles. The profiles are compatible with the inputs of current photochemical air quality models, including the Community Multiscale Air Quality Aerosol Module Version 6 (AE6). The composition of light-duty gasoline PM_2.5 emissions differs significantly between cold start and hot stabilized running emissions, and between older and newer vehicles, reflecting both impacts of aging/deterioration and changes in vehicle technology. Fleet-average PM_2.5 profiles are estimated for cold start and hot stabilized running emission processes. Fleet-average profiles are calculated to include emissions from deteriorated high-emitting vehicles that are expected to continue to contribute disproportionately to the fleet-wide PM_2.5 emissions into the future. The profiles are calculated using a weighted average of the PM_2.5 composition according to the contribution of PM_2.5 emissions from each class of vehicles in the on-road gasoline fleet in the Kansas City Metropolitan Statistical Area. The paper introduces methods to exclude insignificant measurements, correct for organic carbon positive artifact, and control for contamination from the testing infrastructure in developing speciation profiles. The uncertainty of the PM_2.5 species fraction in each profile is quantified using sampling survey analysis methods. The primary use of the profiles is to develop PM_2.5 emissions inventories for the United States, but the profiles may also be used in source apportionment, atmospheric modeling, and exposure assessment, and as a basis for light-duty gasoline emission profiles for countries with limited data.

Implications: 

PM_2.5 speciation profiles were developed from a large sample of light-duty gasoline vehicles tested in the Kansas City area. Separate PM_2.5 profiles represent cold start and hot stabilized running emission processes to distinguish important differences in chemical composition. Statistical analysis was used to construct profiles that represent PM_2.5 emissions from the U.S. vehicle fleet based on vehicles tested from the 2005 calendar year Kansas City metropolitan area. The profiles have been incorporated into the EPA MOVES emissions model, as well as the EPA SPECIATE database, to improve emission inventories and provide the PM_2.5 chemical characterization needed by CMAQv5.0 for atmospheric chemistry modeling.

Background

Composition of PM_2.5 from light-duty gasoline vehicle exhaust

Particulate matter (PM) exhaust from light-duty gasoline vehicles (LDGVs) is a complex mixture composed of elemental carbon (EC), organic carbon (OC), and inorganic elements and ions. Exhaust measurements of particulate mass with aerodynamic diameter less than 2.5 μm (PM_2.5) capture the vast majority of particulate mass emitted from light-duty gasoline vehicles (Robert et al., 2007).

Thermal-optical measurements quantify elemental carbon (EC) that forms from fuel pyrolysis during combustion. Quantifying the EC component of gasoline particulate matter is important for understanding the health risks and the climate-forcing potential of LDGV emissions (EPA, 2012). Additionally, the EC fraction is an important source indicator in source-apportionment studies (Lough et al., 2007; Fujita et al., 2007). Differences in the elemental carbon measurement can have a large impact on the apportionment of emissions between diesel and gasoline engines (Fujita et al., 2007).

In motor vehicle exhaust, the organic carbon (OC) refers to carbon bound with other noncarbon elements in compounds that form liquid and solid aerosols. For our purposes, organic carbon particulate is operationally defined from thermal-optical methods. We refer to organic matter (OM) as the mass of OC and the mass of the associated elements bound to the organic carbon (oxygen, hydrogen, nitrogen, and other noncarbon elements). For light-duty vehicle gasoline exhaust, previous researchers have adjusted OC emission measurements to obtain OM measurements using the following factors: 1.2 (Zielinska et al., 2004; Schauer et al., 2008), 1.25 (Reff et al., 2009), 1.4 (Lough et al., 2005; Robert et al., 2007), and 1.55 (Cheung et al., 2009). Primary OM emissions occur directly from the tail pipe, and include important air toxics such as polycyclic aromatic hydrocarbons (PAHs). Secondary organic aerosol forms from volatile compounds reacting and condensing in the atmosphere (EPA, 2012). Historically, organic matter has been considered the dominant component of primary PM emissions from light-duty gasoline vehicles (Zielinska et al., 2004; Fujita et al., 2007; Lough et al., 2007).

Particulate matter from LDGVs also includes inorganic components, including metals and other elements that contribute significantly to both the mass and the toxicity of PM (Cheung et al., 2010). Light-duty gasoline tail-pipe exhaust is a significant contributor of magnesium, calcium, lead, zinc, molybdenum, and platinum in the roadside environment (Schauer et al., 2006). Similarly, Cho et al. (2009) and Hays et al. (2011) reported that near-road PM_2.5 samples are enriched in metals associated with vehicle emissions, including copper and zinc. Identifying the contribution of each species to LDGV particulate matter is important for understanding the toxicity of the emissions, as well as for estimating the contribution of LDGV exhaust to ambient particulate matter pollution.

Modeling requirements

The Motor Vehicle Emission Simulator (MOVES) is the vehicle emission model developed by the U.S. Environmental Protection Agency (EPA) Office of Transportation and Air Quality to estimate emission rates and emission inventories from on-highway vehicles. MOVES has a range of policy uses, including developing mobile source emission inventories for state implementation plans as required by the Clean Air Act, and developing project-level emission inventories from proposed transportation projects. MOVES is used in conjunction with SMOKE (Sparse Matrix Operator Kernel Emissions) to provide model-ready emission inventories to be used in the Community Multiscale Air Quality System (CMAQ) photochemical air quality model. The EPA uses the MOVES-SMOKE-CMAQ modeling system to evaluate regional and nationwide air quality impacts.

Previous versions of MOVES (2010b, and earlier) estimated PM_2.5 emission inventories, but the model outputs were not aligned with the PM_2.5 model species used by CMAQ. Significant postprocessing of MOVES PM_2.5 outputs was needed to provide PM_2.5 emissions inventories that could be transformed by SMOKE into model-ready inputs of speciated PM_2.5 for CMAQ. The upcoming version of MOVES is designed to produce all PM_2.5 species required by CMAQv5.0, which uses the CMAQ Aerosol Module, version 6, or “AE6.” AE6 includes 18 PM_2.5 species as outlined in Table 1 (Simon, 2012). These additional PM_2.5 species produced by MOVES are beneficial to air quality agencies and researchers who use different air quality models, as they are compatible with previous versions of CMAQ and with the Comprehensive Air Quality Model with Extensions (CAMx, 2011) as shown in Table 1. The PM_2.5 profiles used by MOVES and CMAQ are stored in the EPA SPECIATE database, which is a publicly available repository for PM and VOC speciation profiles (Hsu et al., 2006).

Table 1. PM_2.5 species included in CMAQv5.0 (Aerosol Module 6), CMAQv4.7.1 (Aerosol Module 5), and CAMx5.4

PM2.5 species	CMAQv5.0 species name	Required in CMAQv4.7.1	Required in CAMx5.4
Primary organic carbon	POC	x	x
Elemental carbon	PEC	x	x
Sulfate	PSO4	x	x
Nitrate	PNO3	x	x
Ammonium	PNH4	x	x
Noncarbon organic matter	PNCOM		x
Iron	PFE
Aluminum	PAL
Silica	PSI
Titanium	PTI
Calcium	PCA
Magnesium	PMG
Potassium	PK
Manganese	PMN
Sodium	PNA		x
Chloride	PCL		x
Water	PH2O		x
Primary unspeciated PM2.5	PMOTHR	x	x

Note: Sources: Simon (2012) and CAMx (2011).

Data: Kansas City Light-Duty Vehicle Emission Study

The Kansas City Light-Duty Vehicle Emission Study (KCVES) was conducted to improve PM emission inventories from light-duty gasoline vehicles (Fulper et al., 2010; EPA, 2008). The KCVES provided information on ambient temperature effects on emissions (Nam et al., 2010) and PM emission rates of smoking vehicles (Nam et al., 2008), and it is the major source of light-duty gasoline PM rates in the current MOVES model (EPA, 2011). Several follow-up analyses on the KCVES were conducted to evaluate the effect of sampling temperature (Fujita et al., 2006b) and operating mode on organic speciation (Herrington et al., 2012), chemical analysis of the lubrication oil (Fujita et al., 2006a), and the contribution of lubricating oil to PM emissions (Sonntag et al., 2012).

The KCVES collected PM_2.5 measurements from a statistically representative sample of LDGVs in the Kansas City Metropolitan Statistical Area (KCMSA). The study was conducted in the summer of 2004 (Phase 1) and winter of 2004/2005 (Phase 2). A stratified sampling design was implemented to test a greater percentage of older, high-emitting vehicles compared to the amount of newer, low-emitting vehicles in the U.S. fleet. The LDGVs were grouped into eight strata classified by two vehicle types and four model year groups, as shown in Table 2. The percentages of LDGVs in each strata in the KCMSA are provided in Table 2, as well as the calculated contribution to annual vehicle miles traveled (VMT). Four hundred and ninety-two vehicles were tested across two rounds of the study (42 vehicles were tested in both the summer and winter round). More than 555 vehicle tests were conducted when counting repeat vehicle tests within and across seasons. A subset of 98 vehicles was selected for detailed chemical characterization and is referred as the speciation subset in Table 2. All vehicles were tested on a three-bag LA-92 driving cycle, also known as the California Unified Cycle. The LA-92 included a cold start (Phase 1), hot stabilized running (Phase 2), and a hot start (Phase 3), covering 9.8 miles representing arterial and freeway driving in Los Angeles, CA. The vehicles were purposely tested at ambient temperatures in a garage with open bay doors. The temperature for the testing facility ranged from 58°F to 96°F in the summer round (average of 76°F) and from 12°F to 74°F (average of 43°F) in the winter round (Fulper et al., 2010).

Table 2. Vehicle sample size in the Kansas City Light Duty Vehicle Emissions Study

Vehicle type ^a	Strata	Model year group	Percent of KC LDGV vehicle population ^a	Percent of KC LDGV vehicle miles traveled (VMT)^b		Summer round sample		Winter round sample
						Full sample	Speciation subset	Full sample	Speciation subset
Truck	1	Pre-1981	1.1%	0.6%		2	2	10	3
	2	1981–1990	3.7%	2.4%		21	4	33	3
	3	1991–1995	7.2%	6.5%		18	6	33	7
	4	1996–2005	28.6%	34.2%		39	8	59	11
Car	5	Pre-1981	1.3%	0.7%		6	5	17	3
	6	1981–1890	7.4%	4.6%		49	4	40	5
	7	91–1995	13.4%	11.2%		39	6	44	9
	8	1996–2005	37.3%	39.8%		87	14	41	9
		Sum =	100%	100%		261	49	277	50

Notes: ^aThe definition for car and truck, and the population percentages, are estimated from the 2004 Kansas City Travel Behavior Survey (U.S. EPA, 2008): Passenger cars are defined as coups, sedans, and wagons, trucks are defined as minivans, sport-utility vehicles, and pickups.

^bThe VMT contribution by strata is calculated based on the population percentages combined with annual VMT by vehicle age estimated from the 2001 National Household Travel Survey (NHTSA, 2006).

The emissions measurements made on the KCVES vehicles are grouped into three levels of detail as outlined in Table 3. PM_2.5 measurement data collected at all three levels are used in the profile developed in this paper. At Level 1, each of the 496 vehicles was tested for criteria pollutant emissions—hydrocarbons (HC), carbon monoxide (CO), carbon dioxide (CO₂), oxides of nitrogen (NOx)—and both filter-based PM and indirect measurements of PM. Particulate matter emissions from each phase of the LA-92 were collected on 47-mm Teflo Teflon membrane filters from exhaust extracted from the dilution tunnel and through a PM_2.5 cyclone. Included among the indirect PM instruments is a photoacoustic sampler (DRI, Reno, NV) which measured light-absorbing carbon reported as black carbon (BC), providing a comparable measure to elemental carbon for each vehicle test in the full sample. Details on the DRI photoacoustic sampler are provided in Table S1 in the supporting information.

Table 3. Levels of PM characterization conducted on vehicles tested in Kansas City

	Level 1: Full sample	Level 2: Speciation subsample	Level 3: Speciation composites
Vehicles	496 vehicles	98 vehicles	98 vehicles (same as Level 2)
Measurements	<555 vehicle tests × 3 bags	102 vehicle tests × 3 bags	52 composite samples
Individual vehicle tests	Yes	Yes	Individual vehicle samples for pre-1990 model years vehicles;
			sample composites of one to five vehicles for 1990 and newer model years vehicles
Phase (Bag)-specific measurements?	Yes	Yes	No (composite average of 3-bag LA-92)
Measurements	Gravimetric mass (Teflon filters), CO2, CO, HC, NOx; BC (photoacoustic analyzer); indirect continuous PM (DustTrak, DataRam, quartz crystal microbalance)	Elemental carbon, organic carbon (TOR-IMPROVE, Quartz filters); 49 elements (XRF, Teflon filters)	Sulfate and nitrate (ion chromatography from the sample extracted from the quartz filters). Seven elements (ICP/MS from the sample extracted from the Teflon filters); 293 particle-phase and semivolatile organic species (separate sample train GC/MS, TIGF/XAD).

As summarized in Table 3, at level 2 the 98 vehicles selected for chemical analysis received additional phase-specific measurements of elemental carbon, organic carbon, and metals. One vehicle (1979 Buick LeSabre) was tested in both the summer and winter, yielding 49 and 50 vehicles tested in summer and winter, respectively. EC and OC were measured on quartz-fiber filters for each phase of the LA-92 (Phase 1, Phase 2, and Phase 3, corresponding to Bag 1, Bag 2, and Bag 3), using the thermal optical reflectance (TOR) method with the IMPROVE (Interagency Monitoring of Protected Visual Environments) procedures (IMPROVE-TOR) (Chow et al., 1993). Forty elements were analyzed using x-ray fluorescence (XRF) on the Teflon filters measured on each phase of LA-92.

At Level 3, the vehicle tests measured in Level 2 were composited for measurements of organic compounds and select elements and ions. The composite measurements were not sampled on the same media, but were extracted from filters measured in each bag of the LA-92. Seven elements (Pb, Hg, As, Cr, Cu, Zn, and Mn) were measured by inductively coupled plasma–mass spectrometry (ICP-MS) from the composited Teflon-filter extraction. The elements measured by ICP-MS were previously measured by XRF (Level 2), but ICP-MS has lower detection limits than XRF. Sulfate and nitrate ions were measured with ion chromatography (IC) from the composited extracts from each half of the quartz-fiber filters sampled. Quantification of ammonium and other cations was not budgeted in the study.

In the case of the 1990 and newer model year vehicles, sample was composited from several vehicles (one to five) before chemical analysis. The composting across vehicle tests reduced the 102 individual vehicle tests selected in Level 2 to 52 composite chemical samples. The compositing occurred across several vehicle tests for the newer vehicles in order to assure enough particulate to conduct organic speciation by gas chromatography and mass spectrometry (GC/MS). The semivolatile organic compounds (SVOC) were sampled using a separate sampling train collected with Teflon-impregnated glass fiber (TIGF) filters and XAD cartridge resin, results from which are reported in EPA (2008).

Comparable composite emission rates for PM_2.5 gravimetric mass, EC/OC, and XRF elements was calculated from the individual phase-specific and vehicle-specific measurements, with propagation of uncertainty as described in EPA (2008). Additional details of the vehicle emission testing, sample design, instrument setup, and testing protocol are described in Fulper et al. (2010) and EPA (2008).

Profile Development

Data preparation

To develop PM_2.5 speciation profiles from the KCVES, the total PM_2.5 mass must be reconstructed from the individual particulate species. Several quality control steps are conducted to assure the validity of the individual species measurements: (1) identification of measurements above the detection limit and uncertainty of the analytical measurement; (2) evaluation of the data for inconsistencies or measurement problems; and (3) comparison of the mass balance of individual species to the gravimetric filter mass.

Identification of Significant Measurements

The significance of each PM_2.5 species is evaluated using the composite emission rates for each compound measured in the KCVES (EPA, 2008). The field-blank corrected measurements and their associated measurement uncertainties were calculated in the KCVES report (EPA, 2008) for each composite sample using the minimum detection limit, the precision of the analyte measurement, the concentration of the analyte measurement, and the standard deviation of field-blank measurements. All field blanks were used except the outliers identified from comparisons to the dilution tunnel and exhaust measurements. The uncertainty of the ICP-MS and IC measurements is the uncertainty of the single field-blank corrected composite measurements. For the IMPROVE-TOR and XRF measurements, the uncertainty measurements of the composite sample is calculated by propagating the uncertainty from the individual filter measurements from each phase of the LA-92 drive cycle and from each vehicle test. The steps and equations used to calculate the composite measurement uncertainties are provided in the KCVES report (EPA, 2008).

In this work, we conducted a statistical test to determine if the measured compounds were significantly above the dilution tunnel blank measurements. First, we calculated the dilution-tunnel corrected composite values and propagated the uncertainty of the composite measurements with the standard deviation of the dilution-tunnel blank measurements taken in each season. Six composite dilution tunnel blanks were calculated from each round of the study from dilution tunnel blank tests made almost on a day-by-day basis (EPA 2008). Second, we counted the occurances of measurements made above the dilution-tunnel corrected uncertainty for the EC, OC, SO₄, NO₃, 49 XRF, and 7 ICP-MS elements measured in all 52 composite samples. Third, we calculated the probability of observing the counted occurences due to the variability of the measurement alone using the binomial distribution. The binomial distribution is used because the measurement uncertainty calculated in the KCVES report (EPA, 2008) varies with each composite sample result. Results are summarized in Table 4 for each compound measured in the composite sample with the corresponding number of measurements that were made above the dilution-tunnel corrected uncertainties, and the probability of observing that number of occurances due to the measurement uncertainty alone (referred to as type I error).

Table 4. Significance test results of species measurements from the KCVES program

Type I error	Measurements observed above the measurement uncertainty	Species
3.0E-15	52	PM mass, EC, OC Ca, Fe, S, Zn, SO4 Cr-ICP, Mn-ICP, Zn-ICP Pb-ICP
3.0E-15	51	Si
3.0E-15	50	Cu-ICP
3.0E-15	48	Cu
3.0E-15	47	P
3.0E-15	39	Cr
3.0E-15	35	Cl	 ← Significant →
4.3E-15	33	Ni, Pb
1.6E-14	32	Br
2.0E-09	26	K
5.2E-06	21	Al, NO3
7.1E-05	19	Na
0.0021	16	As-ICP
0.0054	15	Mo
0.013	14	Mn
0.059	12	Ti
0.31	9	Mg, Zr, Hg-ICP
0.74	6	Ba	← Insignif cant →
0.85	5	Ga
0.93	4	Sr
0.97	3	Sn, Tl
0.99	2	V, As, Rb, Pd, Ag, Hg
1.00	1	Se, Y, Sb, La, Au, U
1.00	0	Co, Cd, In

Note: Elements measurements are made from XRF (x-ray fluorescence) unless noted by ICP-MS (inductively coupled plasma–mass spectrometry).

For example, titanium was measured above the dilution-tunnel corrected measurement uncertainties on 12 of the 52 composite samples. Assuming the measurement uncertainty is normally distributed, the probability of observing 12 or more measurements above the uncertainty is equal to 0.06. Titanium (Ti) and manganese (Mn) could be considered significant depending on the acceptable type I error and which correction is used (dilution-tunnel corrected, or field blank only). Magnesium (Mg), zirconium (Zr), and mercury-ICP (Hg-ICP) each had only nine composite measurements above the uncertainty of the dilution-tunnel corrected values. These and the elements with fewer measurements above the uncertainty are considered insignificant using reasonable type I error values. Table 4 identifies the species according to significant and insignificant based on a type I error of 0.10. Only the elements that are deemed significant (titanium and above) are included in the PM_2.5 speciation profile. Additional details on dilution tunnel-correction and the statistical test are included in the supporting information.

Inorganic species

The PM_2.5 profile is not intended to be an exhaustive list of elements and ions that are emitted in the particle phase, but is intended to capture the species that contribute significantly to the total PM_2.5 mass and/or are required by CMAQv5.0. As stated, 10 of the 12 additional PM_2.5 species required by CMAQv5.0 are elements. Elements and ions are included in the PM_2.5 profile from the KCVES if they are valid measurements. Other sources are used to impute element and ion measurements if no valid measurements from KCVES are available and they are required by CMAQv5.0 AE6 and/or for achieving mass balance.

Several elements known to be emitted in light-duty gasoline PM_2.5 are not included in the profile, including trace elements in the catalytic converter washcoat (cesium, rhodium, palladium, platinum, lanthanium), which deteriorate and are emitted in light-duty exhaust (Van Der Schoot et al., 2001). These elements were either not measured or were not detected at significant levels in the KCVES. The excluded elements are emitted in such small concentrations that they do not impact the mass balance of filter mass, nor are they required by CMAQv5.0 AE6. The arsenic measurements are also excluded from the species profile. Although the arsenic ICP-MS measurements were deemed significant, the measurements exhibited strong inconsistencies across the study. The differences in the XRF and ICP-MS measurements for arsenic were an order of magnitude, and more than half of the arsenic ICP-MS values in the winter round were negative. Arsenic is not required by CMAQv5.0 AE6 and values are not imputed from other studies.

In most cases the elements that were measured by both XRF and ICP-MS were within the statistical variability of each other. Because XRF measurements were conducted at the bag level, the XRF measurements are used in the reconstructed PM_2.5 profile for chromium, zinc, lead, and copper. In the case of manganese, the ICP-MS and XRF measurements were highly correlated, but a significant difference is detected. The bag-level XRF measurements of Mn were thus scaled to the ICP-MS measurements for use in the PM_2.5 profile. Sodium (Na) and magnesium (Mg) measured by XRF are deemed qualitative measurements because they emit weak x-rays due to their small size (David Campbell, personal correspondence). Additionally, the XRF measurements of magnesium were not significantly above the uncertainty of the dilution-tunnel measurements. No measurements of sodium or magnesium were made by the ICP-MS in the KCVES. To provide quantitative emission rates for sodium and magnesium, values are imputed from measurements made using ICP-MS from 53 light-duty gasoline vehicles in the South Coast Air Quality Management District in California (Schauer et al., 2006; Lough et al., 2007). The major source of sodium and magnesium is believed to be the engine oil since both of these were measured at significant levels in the used oil from a subsample of vehicles tested in the KCVES. As such, the emissions of sodium and magnesium are imputed using the ratio of sodium and magnesium to zinc (1:15 and 1:4.1, respectively). Zinc is selected because it is also a component of lubricating oil, and because the ratio of Zn to PM_2.5 is roughly equal between the summer testing of KCVES and the profile used in the Schauer et al. (2006) study.

Silicon was the most abundant element measured in KCVES (EPA, 2008). The catalytic converter washcoats, substrates, and mats are composed of materials containing silicon (including silica, alumina silicate, vermiculite, and cordierite) (Heck et al., 2002). Erosion of these materials in the exhaust system (Watson and Dietz, 2009) could lead to comparable emissions of both silicon and aluminum in the exhaust (Schauer et al., 2002). However, in KCVES, the silicon emissions were much higher than the aluminum emission rates. Schauer et al. (2002) reported silicon emission rates from nine catalyst vehicles ranging from 1981 to 1994 model years that were one order magnitude smaller than in KCVES. No silicon was detected in the used lubricating oil in the KCVES (Fujita et al., 2006a), indicating that the large silicon emission rates were not coming from the engine or the emission control system.

Upon further review, we concluded that silicone rubber in the KCVES sampling system contaminated the particulate measurements with silicon and other organic material. In the KCVES sampling system, silicone rubber boots were used as a coupler between the exhaust tail pipe of the recruited vehicles and a stainless steel exhaust transfer line. In a number of the KCVES tests, several inches of the silicon rubber boots were directly exposed to the hot exhaust. Silicone polymer is known to volatize and breakdown when exposed to high temperatures and can be oxidized from lubricating oil (Keller, 1986). Cadle et al. (1999) suspected that silicone couplers may have contributed to particulate measurement artifacts in a past LDGV emission study, and Maricq et al. (1999) confirmed the silicone couplers can contribute to high concentrations of ultrafine particles measured under high exhaust temperatures.

The silicone contribution to the measured PM_2.5 mass is estimated using the elemental silicon measurements. The silicone polymer contains silicon, oxygen, carbon, and hydrogen, which contribute to the measured particulate and organic carbon mass; however, the chemical structure of the volatized silicone material measured on the measurement filters is unknown. To estimate the contribution of volatized silicone relative to the elemental silicon measurements, we used a regression model developed in an earlier analysis (Sonntag et al., 2012) to estimate the relative contribution of silicon to the measured PM_2.5 emissions. The coefficient of silicon was statistically significant for the linear regression of pre-1990 vehicles, with a mean value of 4.075 with 95% confidence intervals of (2.47 to 5.67). Thus, the actual contribution of the volatilized silicone to the particulate matter measurements is estimated to be 4.075 times the measured elemental silicon. The silicon coefficient for predicting PM_2.5 for newer vehicles (1990–2004) is 3.84 (Table S4), which is not statistically significant but is quite similar to the coefficient derived for the pre-1990 vehicles. Using the 4.075 silicone mass ratio and the measured elemental silicon from the speciated subset vehicles in KCVES, the volatized silicone contribution to measured PM_2.5 is estimated for each sample, as shown later in Figures 2 and 4.

Figure 1. Comparison of mass-balanced corrected organic matter emission rates with measured organic carbon emission rates.

Figure 2. Average reconstructed PM_2.5 composition of cold start emissions.

Elemental silicon values are model species in CMAQv5.0 AE6. Because the measured elemental silicon is contaminated by the silicone rubber housing, the silicon emission rates for the profile are substituted with values equal to the KCVES aluminum emission rates for both the cold start and Bag 2 emissions. Aluminium is used due to its similar sources in the emission control system and similar measured emission rates reported from previous LDGV studies not using silicon couplers in the sampling system (Schauer et al., 2002)

The sulfate ion (SO₄) and elemental sulfur (S) emission rates for the composite samples are compared in Figure S5 as well as in EPA (2008). The two measurements are highly correlated (r² = 0.97), and the IC measured sulfate accounts for 81% of the elemental sulfur mass measured by XRF. Lack of mass balance could be due to both (1) measurement uncertainties, including different capture efficiencies of sulfur using the different media (Teflon and quartz) and differences in the XRF and IC measurement methods, and/or (2) particulate sulfur existing in different molecular states such as zinc dithiophosphate (a lubricant additive). Trace amounts of sulfur-containing PAHs were measured in the study (benzonaphthothiophene and dibenzothiophene), which emitted typically less than 0.5% of the sulfur measured by XRF. To estimate the amount of sulfate at the vehicle and bag level, the composite SO₄ measurements are apportioned using the XRF S measurements and the SO₄ to S ratio from each composite sample. No estimate of nonsulfate particulate sulfur is included in the reconstructed mass.

Nitrate is measured at the composite level, and is apportioned for individual vehicles and each bag of the LA-92 using the nitrate/PM_2.5 ratio from each composite test and the PM_2.5 measurements from each vehicle and bag. Ammonium concentrations were estimated in the absence of measurement data. Chueng et al. (2009) estimated ammonium emissions on gasoline vehicles 12–20 times higher than from diesel vehicles. Previous test measurements of ammonium emissions from light-duty gasoline PM have found ammonium concentrations in stoichiometric equilibrium with measurements of sulfate and nitrate, in the form of ammonium sulfate and ammonium nitrate (Fujita et al., 2007, Schauer et al., 2006, 2008). The source of the filter-measured ammonium in these studies may be the reaction of the sulfate ion and ammonia gas in the filters (Schauer, 2008). Because filter mass is used as the reference for estimating the particulate matter composition in this study, ammonium is imputed assuming the sulfate and nitrate is emitted as ammonium sulfate and ammonium nitrate.

Metal-bound oxygen

In comparing the element measurements to the filter mass, we accounted for the metal-bound oxygen by using the oxide state of each metal shown in Table 5. Calcium, zinc, and magnesium are assumed to be in the form of phosphates and sulfates, which is consistent with chemical analyses of light-duty gasoline particles (Sodeman et al., 2005), light-duty catalytic converters (O’Neill et al., 2006), and diesel particulate ash (Sappock, 2009). In Table 5, calcium, zinc, and magnesium are not listed in oxide states, because the oxygen is associated with the phosphate and sulfates. Common oxide states used for other elements are taken from Reff et al. (2009), including averages where several common oxide states are assumed. Chlorine and bromine are not assumed to be bound to oxygen.

Table 5. Oxide states assumed for calculation of metal-bound oxygen

Element	Oxide form 1	Oxide form 2	Oxide form 3	Oxide/element mass ratio
Na	Na2O			1.35
Mg	Mg			1.0
Al	Al2O3			1.89
Si	SiO2			2.14
P	PO4			3.07
Cl	Cl			1.0
K	K20			1.20
Ca	Ca			1.0
Ti	TiO2			1.67
Cr	Cr2O3	CrO3		1.69
Mn	MnO	MnO2	Mn2O7	1.63
Fe	FeO	Fe2O3		1.36
Ni	NiO			1.27
Cu	CuO			1.25
Zn	Zn			1.0
Br	Br			1.0
Mo	MoO2	MoO3		1.42
Pb	PbO	PbO2		1.12

Elemental carbon

IMPROVE-TOR elemental carbon measurements were made on each of the vehicles tested in the speciation sample. Because all vehicles had BC measurements with the photoacoustic instrument, this subset of vehicles is used to compare the TOR-EC and photoacoustic BC measurements (EPA, 2008). For conventional diesel exhaust, the two instruments provide similar magnitude of emissions (Moosmüller et al., 2001). In the KCVES, the two instruments provided close to 1:1 agreement for the heavily loaded bag 1 samples, but exhibited poor correlation for the lightly loaded bag 2 and bag 3 samples (Figures S7–S9). The difference in the BC and TOR-EC emission rates on the lightly loaded PM samples is attributed to the adsorption of organic compounds on the quartz filters (EPA, 2008).

The BC measurements from the photoacoustic instrument are used in the PM_2.5 profile instead of elemental carbon since these were less affected by positive artifact at low emission levels and were measured on all 496 vehicles. In the final PM_2.5 profile, the BC measurements reported from the photoacoustic instrument are used to estimate the CMAQv5.0 AE6 definition of EC (Table 7, shown later). The BC are not adjusted to represent EC because of the close 1:1 agreement observed in the bag 1 samples (Figure S7).

Organic matter

To account for positive measurement artifact of the organic carbon and the noncarbon organic material, the OM emission rates can be estimated using the following equation:

(1)

where OM is organic matter, OC is organic carbon, k is the factor used to account for the noncarbon material associated with the carbon, and b is the correction for measurement artifacts. Quartz-fiber filters are known to be susceptible to positive measurement artifacts for organic carbon, which occur when volatile and semivolatile organic vapors adsorb onto the filter (Noll and Birch, 2008). The extent of organic artifact on quartz-fiber filters depends on multiple factors, including the PM loading of the sample, the hydrocarbon emissions, and other sampling conditions including dilution and pressure drop across the filter (Matti et al., 2011).

A backup quartz-fiber filter placed in series behind the primary filter can be used to estimate the positive organic artifact (b in eq 1). However, a backup quartz-fiber filter was only measured on a subset of 10 vehicles sampled during the winter phase of the KCVES (Fujita et al., 2006b). Additionally, using a backup filter correction assumes that the adsorption of the vapors is equal on both the primary and secondary filters and can be problematic for low-emitting vehicles where the OC measurements can be as large on the backup filter as on the primary filters (Fujita et al., 2006b). Also, because the volatilized silicone rubber from the sampling system likely contributed to the primary OC measurements, a backup filter would not account for the silicone measurement artifact.

Rather than using backup filter correction of the OC measurements, we used a mass-balance approach to estimate the OM emission rates. This approach enables the artifact correction and silicone contamination correction to be estimated from data collected on each test. Mass balance approaches have been used to estimate atmospheric organic carbon concentrations (Frank, 2006; El-Zanan et al., 2009) and were previously used by the EPA in the postprocessing of MOVES on-highway PM_2.5 emission inventories for air quality modeling (Strum, 2011). The organic matter emissions were estimated for each test by the difference of the measured total mass and the individual components, as shown in eq 2:

(2)

The OM emission rates estimated from the mass balance approach are compared to the measured organic carbon emission rates in Figure 1. Simple linear regression models are fit to the data. In each case, the intercept is negative, indicating a positive artifact in the measurements as proposed in eq 1. The slope of the regression ranges from 1.19 to 1.24, which corresponds well with the range of OM to OC ratios reported in the Background section.

Figure 3. Average PM_2.5 fraction of each measured species for cold start emissions by model-year group, accompanying 95% confidence intervals.

As shown in Figure 1, the mass-balance approach occasionally yields either negative estimates of OM or OM estimates that are much larger than the original OC measurements. Nineteen of the 270 estimated OM values are negative. Seven of the negative OM estimates resulted from the photoacoustic measurements being larger than the elemental carbon or gravimetric filter mass measurements. The photoacoustic measurements are used exclusively rather than the IMPROVE-EC values in the final PM speciation profile. However, in cases where (1) the reconstructed PM mass is greater than the measured filter mass and (2) the photoacoustic measurements are larger than the TOR-EC measurements, the TOR-EC measurements are used in eq 2 to estimate positive OM emission rates. The remaining 12 negative OM values were imputed with zero values. To avoid OM rates being biased high from the mass-balance approach, the OM estimated values are capped at twice the measured OC emission rates, which impacted 23 of the 270 initially estimated OM values. A factor of 2 is selected because it is the largest OM/OC ratio reported in a range of combustion sources in Reff et al. (2009).

The estimated ratio of estimated OM to measured OC by bag of the LA-92 and model year group is displayed in Table 6. In general, the OM/OC ratio is greater than 1, which is expected due to the additional mass of the noncarbon atoms in the organic matter. The estimated ratio is less than one for some model year groups in Bag 3, indicating that much of the organic carbon measured in Bag 3 is considered a positive artifact as defined in this paper. Contribution of silicone to the measured OC could also be another reason many of the ratios in Table 6 are less than 1.2, especially for the older vehicles.

Table 6. Mass-balance OM compared to measured OC

Bag	Model group	n	Ratio of mass-balance OM to measured OC	Standard error	Lower 95% confidence interval	Upper 95% confidence interval
Bag 1	Pre-1981	14	1.16	0.04	1.07	1.26
	1981–1990	17	1.25	0.06	1.13	1.37
	1991–1995	27	1.17	0.08	1.01	1.34
	1996–2005	37	1.09	0.09	0.90	1.27
Bag 2	Pre-1981	14	1.05	0.15	0.74	1.37
	1981–1990	15	1.14	0.07	1.00	1.28
	1991–1995	28	1.22	0.09	1.03	1.40
	1996–2005	42	1.20	0.08	1.03	1.37
Bag 3	Pre-1981	12	0.98	0.29	0.35	1.61
	1981–1990	14	1.16	0.09	0.97	1.36
	1991–1995	22	0.60	0.12	0.36	0.84
	1996–2005	27	0.73	0.14	0.44	1.01

Profile analysis

Average PM_2.5 profiles are estimated for cold start and hot stabilized running processes using the bag-specific PM_2.5 reconstructed data. Running emissions in MOVES are defined as emissions from the vehicle when the engine block is warm and the operating catalyst temperature is reached, which we refer to as hot stabilized running. We use data from Bag 2 of the LA-92 cycle for the hot stabilized running speciation profile. Cold start emissions are typically defined as the incremental emissions produced from the initial cold start before the engine (and catalyst) have fully warmed. The PM_2.5 emission rates for cold start in MOVES are based on the difference of Bag 1 and Bag 3 (warm start) emissions, which have identical speed traces. The cold start PM_2.5 emission profile is similarly estimated by subtracting Bag 3 emissions from Bag 1 emissions for each of the measured species from the mass-balanced profiles. In modeling the cold start as Bag 1 – Bag 3, we assume that (1) all of the incremental cold start emissions are captured by Bag 1, and (2) the warm start is a minor contributor in comparison to the Bag 3 running emissions. In reality, a significant temperature dependence is observed in Bag 2 PM_2.5 emissions (Figure S18), meaning that the LA-92 Bag 1 is likely too short a cycle to fully warm the vehicles to a stable hot stabilized running condition. The persistent temperature effect on the Bag 2 is considered in the development of the PM_2.5 profile.

We developed the PM_2.5 profiles using the largest sample size of measurements available. The BC fraction of PM_2.5 is estimated using data from the full test program (496 vehicles) because photoacoustic and gravimetric filter measurements were made for each test. The data set includes 42 vehicles that were tested in both the summer and winter. For statistical analysis, the 42 vehicles are treated as separate vehicles in the summer and winter because they compose a small fraction of the total tests and because the large temperature impacts on emissions are anticipated to dominate inter-vehicle variability. For estimating BC/ PM_2.5, tests are excluded that did not have detectable BC or PM_2.5 emissions. For estimating the hot stabilized running emission rates on Bag 2, the sample size is reduced from 538 to 517 tests. For cold starts, additional tests are excluded when the BC or PM_2.5 emissions on Bag 3 is greater than on Bag 1, which reduced the data set to 473 tests (Table S2).

The speciation sample (102 tests) is used to create the speciation profile for the composition of the organic matter, elements, and ions. Ninety-nine of the 102 speciation tests on Bag 2 have valid PM_2.5, EC, and OC measurements for development of the hot stabilized running profile (Table S3). For cold starts, only 66 out of 102 speciation tests are deemed valid where roughly one-third of the samples are excluded due to negligible PM_2.5, EC, and OC readings on calculated differenced values (Bag 1 – Bag 3). Due to the large number of excluded samples, the Bag 1 PM_2.5 emissions of the excluded cold start tests were compared with all of the valid bag 1 PM_2.5 emissions in Figure S2. No systematic bias is observed in the reduced or excluded PM_2.5 Bag 1 emision rates, suggested that the reduced sample of the cold start emissions used for the speciation profile is still representative of the larger sample. Speciation factors are estimated for each of the species required by CMAQv5.0 and additional elements measured at significant levels (phosphorous, chromium, nickel, copper, zinc, bromine, and molybdenum).

Cold start emission profile

The reconstructed start PM_2.5 measurements are initially averaged according to the eight vehicle strata and seasons in Figure 2, and are used to evaluate the impact of season, vehicle type and model year group on the PM_2.5 composition. As shown, the cold start profile is dominated by BC and OM emissions, with a very small inorganic contribution. The BC and PM_2.5 cold start emission rates are much larger in the winter than the summer. In the full vehicle sample (473 vehicle tests), the PM_2.5 emissions are larger in winter in each stratum (Figure S10), and the BC emissions are significantly larger in all but the pre-1981 truck strata (Figure S11). The BC/PM_2.5 ratio remains relatively consistent across seasons using data from the full sample (Figure S12), suggesting that the BC and PM_2.5 cold start emissions increase at the same rate at colder temperatures.

The overall emissions of OM, elements, and sulfate emissions also increase significantly at cold temperatures. However, the PM_2.5 fraction of organic matter, sum of elements, and sulfate emissions does not vary significantly between seasons for each of the strata (Figures S14–S16). Herrington et al. (2012) found that the PM_2.5 proportions of key particulate organic classes (hopanes, steranes, and polycyclic aromatic hydrocarbons) were relatively invariant to changes in season in a subsample of 18 vehicles tested in both seasons in the KCVES. Both analyses suggests that while cold start LDGV PM_2.5 emissions are very sensitive to temperature, the relative composition of the cold start PM_2.5 emissions is consistent across temperatures. For the final cold start emission profile, the measurements taken in the summer and the winter are equally weighted.

Figure 2 shows that there are no strong differences in the composition between cars and trucks. This is supported by the BC/PM_2.5 fraction between cars and trucks shown in Figure S12. Figure 3 graphs the cold start PM_2.5 fraction of each measured species averaged by model year groups, to isolate the model-year impact on PM_2.5 speciation. Overall, all PM_2.5 species decrease with new model year vehicles (Figure S17), but important differences are noted in the proportion of each species. The BC fraction appears to increase slightly for the 1990 and later model year groups (Figure S11), but is difficult to determine due to the variability in the data. The data are pooled to two larger model year groups: pre-1991 and the 1991–2005 model year groups (Figure S13), where the increase in BC/PM_2.5 is shown to clearly be significant at the 95% confidence level. Such an increase in the BC/PM_2.5 fraction is consistent with previous research (Robert et al., 2007). The fraction of OM decreases as the overall PM_2.5 is reduced in the newer model year groups. The trend is reversed for the inorganics, where the relative fraction of sulfate and most elements increases with decreases in PM_2.5. Only elements that were measured at significant levels in the cold starts are included in the cold start profile, which leaves 7 of the 14 elements identified as significant in the composite profile (Ca, Fe, P, Zn, Pb, Cr, Ni).

Figure 4. Average reconstructed PM_2.5 composition of Bag 2 emissions.

Hot stabilized running emission profile

Figure 4 plots the reconstructed emission rates averaged by season and strata for the PM_2.5 hot stabilized running emissions. An increase in PM_2.5 emissions in the winter sample is observed in the mean PM_2.5 emissions of the newest model year vehicles (1996–2005) in Figure 4, but no consistent trend is observed for the older model year vehicles. The variability of the PM_2.5 emissions across season is not surprising, given the small sample size of vehicles, and differences in the vehicles tested in the summer and winter rounds. The wintertime increase in PM_2.5 emissions for the newest vehicles appears to occur in the BC and OM fractions. The hot stabilized running BC emissions are evaluated in the supporting information on the full data set vehicles (Figures S19 and S20). While BC emissions increase in the winter, there is no discernible season/temperature trend on the BC/PM_2.5 fraction.

The large contribution of volatized silicone is especially evident in the PM_2.5 composition of the Bag 2 emissions. The trucks have a higher contribution, which is expected due to higher exhaust temperatures and larger exhaust tail pipes, which expose more silicone rubber to the hot exhaust. The summer samples have a larger contribution of silicone. Besides the silicone contribution, the composition of the remaining PM_2.5 is similar between cars and trucks for each model-year group. The only significant difference noted between the speciation profiles for cars and trucks is for the pre-1981 model-year group. The pre-1981 cars had a relatively large BC fraction, and the trucks have a large sulfate fraction, which could be due to higher prevalence of oxidation catalysts in pre-1981 trucks. However, this difference could also be attributed to the small sample size of the oldest model-year group, and large variation of emissions from the oldest and most deteriorated LDGVs, which is expected based on previous studies (Lough et al., 2007; Lough and Schauer, 2007).

To create a speciation profile for emissions modeling for hot stabilized running emissions, the summer and winter measurements are equally weighted, and the silicone contribution to PM is removed. The average hot stabilized running PM composition is significantly different according to model year groups as shown in Figure 5. The fractions of the major components—BC, OM, and SO₄—are relatively unchanged between model-year groups, while the fraction of metals and other minor elements significantly increases with newer model-year vehicles. Each of the 14 elements significant in the composite profile is also significant in the hot stabilized running profile.

Figure 5. Average PM_2.5 fraction of each measured species for hot stabilized running emissions by model-year group, accompanying 95% confidence intervals.

Final PM_2.5 speciation profiles for light-duty gasoline vehicles

We considered the level of aggregation when constructing PM speciation profiles for emission modeling in MOVES. Although the data show model-year differences in the PM composition, we propose using a fleet-average PM profile to capture the prevalence of high emitters within all model year-groups, and to avoid overfitting model year trends. Malfunctioning high-emitting vehicles are known to contribute a significant share of in-use PM emissions from light-duty vehicles (Robert et al., 2007; Carroll et al., 2011; Lough and Schauer, 2007; Cadle et al., 1999). High-emitting gasoline emissions have a highly variable PM composition, due to failed emission control systems, excessive oil consumption, and poor fuel control. Previous analysis of the KCVES suggested that the speciation subsample (102 tests) provides a reasonable estimate of the total PM mass compared to the full sample (522 tests), but the speciation sample underestimated the high emitting vehicles in the newer model-year groups (Sonntag et al., 2012). Other test programs have confirmed that high-emitting gasoline vehicles also occur in modern vehicles, such as 1990s-era vehicles with electronic fuel injection (Robert et al., 2007; Carroll et al., 2011). The speciation sample size was deemed too limited to accurately capture the impact of deteriorated vehicles within each model-year group. Using all the data in a fleet-average approach is deemed more appropriate to characterize representative PM_2.5 composition from in-use light-duty gasoline vehicles.

The fleet-average PM speciation profiles are calculated using seasonal, vehicle miles traveled (VMT), and PM mass weighting. The PM profile is calculated using the ratio of the mean emissions of each PM_2.5 species, also referred to by others as a mass-normalized emission profile (Schauer et al., 2006). The vehicle tests from each season are equally weighted, and the vehicle tests are averaged according to the calculated contribution to annual VMT in the Kansas City MSA (Table 2). The PM species are calculated using the Ratio statement in SAS 9.2 Proc SurveyMeans, which calculates standard errors for the ratios using Taylor series linearization. By using VMT and mass weighting, the profile scales the contribution of older and higher emitting vehicles up according to their high PM emissions, but also scales their contribution down based on their relatively small contribution of VMT.

The OC, elements, and ion emission factors are harmonized with the BC speciation factor to create a single PM speciation profile. First, the BC ratios computed from the full sample are used, because they are based on the largest number of vehicles. Then the speciation factors calculated from the speciation subsample are used to speciate the “non-BC” fraction from the full sample. This caused a 6.17% increase in the non-BC species for cold start emissions and a 0.11% increase in hot stabilized running emissions due to differences in the mean BC fractions measured from the full samples and the speciation samples. The resulting PM speciation factors for the hot stabilized running and cold start emissions are located in Table 7. The OC fraction is back-calculated from the OM fraction assuming a 1.2 organic matter correction. “CMAQ5.0 AE6 unspeciated” is composed of unspeciated filter mass, and species not reported by CMAQ 5.0 AE6 (primarily composed of metal-bound oxygen, and two measured metals not reported by CMAQ: zinc and phosphorous). The 95% confidence intervals are based on the original samples as shown in Table 7. Confidence intervals are also provided for the imputed values but should be considered semiqualitative compared to the measured species.

Table 7. PM_2.5 profile for cold start and hot stabilized running emissions, with weighted average using vehicle miles traveled (VMT)

	Start				Running				Measurement/source
PM species	n	Mean ratio ± 95% CI			n	Mean ratio ± 95% CI
Elemental carbon (EC)	484	443.69	±	42.96	531	140.00	±	26.79	BC measured by photoacoustic analyzer. Used to represent CMAQ5.0 EC.
Organic carbon (OC)	66	426.39	±	66.33	99	557.00	±	40.18	5/6 of organic matter, calculated by mass balance of each sample
Noncarbon organic matter (NCOM)	66	85.28	±	13.27	99	111.40	±	8.04	1/6 of organic matter
SO4	66	9.45	±	2.43	99	71.87	±	19.03	Ion chromatography
NO3	66	2.63	±	0.82	99	2.89	±	0.78	Composite measurements allocated to bag by mass fraction
NH4	66	4.31	±	1.02	99	27.79	±	7.25	Imputed assuming ammonium sulfate and ammonium nitrate
Fe	66	3.07	±	2.14	99	18.28	±	5.26	XRF
Al			±		99	3.24	±	0.98	XRF
Si			±		99	3.24	±	0.98	Imputed from Al based on Schauer et al. (2002)
Ti			±		99	0.34	±	0.14	XRF
Ca	66	3.88	±	1.43	99	14.43	±	2.58	XRF
Mg	66	0.19	±	0.18	99	1.45	±	0.25	Imputed from ratio to Zn based on Schauer et al. (2006)
K			±		99	0.85	±	0.26	XRF
Mn			±		99	0.24	±	0.15	XRF/ICP (composite ICP measurements allocated to bag with XRF)
Na	66	0.05	±	0.05	99	0.39	±	0.07	Imputed from ratio to Zn based on Schauer et al. (2006)
Cl	66	0.21	±	0.12	98	1.04	±	0.42	XRF
P	66	1.51	±	0.51	99	4.34	±	0.79	XRF
Cr	66	0.20	±	0.11	99	0.73	±	0.20	XRF
Ni	66	0.08	±	0.05	99	0.27	±	0.11	XRF
Cu			±		99	0.96	±	0.24	XRF
Zn	66	0.78	±	0.72	99	5.93	±	1.02	XRF
Br			±		99	0.22	±	0.08	XRF
Mo			±		99	0.21	±	0.06	XRF
Pb	66	0.26	±	0.11	99	0.48	±	0.14	XRF
Metal-bound oxygen (MBO)	66	4.06	±	2.33	99	26.80	±	5.20	Assumed metal oxide states
KCVES unspeciated	66	13.95	±	18.22	99	5.63	±	6.77	Remainder
CMAQ5.0 AE6 unspeciated (PMOTHR)	66	20.85	±	17.54	99	45.57	±	10.96	Remainder + elements not included by AE6 CMAQ5.0

Note: Ratio values are in mass per total PM_2.5 (mg/g).

Summary

Important differences between cold start and hot stabilized running emissions are captured in the developed PM profiles. Cold start PM emissions are dominated by BC emissions from incomplete fuel combustion. BC emissions account for 44% of the cold start emissions, but only 14% of the hot stabilized running emission profile. The fraction of OC is relatively constant between the start and running emissions, accounting for 43% of the cold start and 56% of the hot stabilized running emissions. The estimated OC fraction is smaller than previous LDGV test programs (Schauer et al., 2006; Zielinska et al., 2004), which may be due to (1) the positive OC artifact correction accounted for in this study and (2) this profile being weighted heavily toward newer vehicles, which have lower fractions of OC in the total PM emissions.

Important differences between the cold start and hot stabilized running PM_2.5 profiles are evident in the inorganic fractions. The sum of measured elements constitutes 5.6% of the hot stabilized running emission profile, but only 1.0% of the cold start emission profile. Many of the elements that were significant in the LA-92 composite cycles were not significantly different from zero in the cold start emission profile. The dominant elements in the hot stabilized running emission profile are engine and exhaust system wear components (iron, aluminum) and lubricating oil additives (calcium, phosphorous, zinc). Similarly, the inorganic ion fraction is considerably higher in the hot stabilized running emissions (10.3%) compared to the cold start emissions (1.6%).

By incorporating separate PM profiles for cold start and hot stabilized running LDGV emissions, important differences in the composition of primary PM_2.5 emissions can be modeled. The LDGV start emission rates are very sensitive to ambient temperature in the MOVES model, and essentially double with every 15°F drop in ambient temperature, based on previous analysis (Nam et al., 2010), whereas MOVES running emissions are less sensitive to ambient temperature. With separate profiles for start and running operating conditions, different PM composition can be modeled for regions with different cold start and hot stabilized running emission contributions. For example, PM emission inventories in cold regions can model an increased fraction of fuel-derived BC emissions due to a larger contribution from PM start emissions.

The developed profiles provide an important update on PM characterization from a large sample of in-use LDGVs as of 2005 for use in emission modeling. The speciation profile is VMT weighted to be representative of the U.S. on-road vehicle fleet, as represented by the vehicles in the Kansas City Metropolitan Area. The same data could be reweighted to represent vehicle fleets in other areas such as developing countries, where test data are limited but the fleet technology and age distribution can be estimated.

Additional speciation data are needed for estimating the impacts of changing fuels and technologies. For the next generation of vehicles, the composition of PM is expected to become increasingly dominated by BC emissions from both low-emitting port-fuel injected vehicles (Robert et al., 2007; Kleeman et al., 2008; Ayala et al., 2011; Robinson et al., 2013) and gasoline direct injection vehicles (Andersson, 2009; Storey et al., 2012; Forestieri et al., 2013). Additional light-duty gasoline PM profiles should be added to MOVES and SPECIATE as such data becomes available.

Supplemental Data

Supplemental data are available for this paper. Go to the publisher’s online edition of the Journal of the Air & Waste Management Association. Supporting information includes details on quality assurance, details on the statistical test used to identify significant measurements, additional information on quality assurance of element and ion measurements, analysis of the silicone contamination issue, and supporting figures used in developing the PM speciation profile.

Acknowledgment

This work was made possible by the contributions of many individuals and organizations. Contributors at the EPA include Chad Bailey, Joseph Somers, Madeline Strum, Richard Snow, James Faircloth, John Koupal, Edward K. Nam, and Megan Beardsley. Eric Fujita and David Campbell at Desert Research Institute (DRI) provided addition information on the chemical measurements made in KCVES. Sandeep Kishan and Scott Fincher at Eastern Research Group (ERG) provided assistance in obtaining data collected from the KCVES. The KCVES was organized and implemented by ERG, and its subcontractors, DRI, BKI, NuStats, Sensors, and ESP, under the direction of Sandeep Kishan.

Funding

Funding for PM speciation profile development work reported in this paper was provided through the American Association for the Advancement of Science (AAAS) fellowship under Cooperative Agreement X3-83459201 awarded by the EPA to the AAAS, and through the Oak Ridge Institute for Science and Education (ORISE) program funded by the EPA Office of Transportation and Air Quality.

Additional information

Notes on contributors

Darrell B. Sonntag

Darrell B. Sonntag and Carl R. Fulper are engineers, and Catherine A. Yanca is an environmental scientist at the National Vehicle Fuel Emissions Laboratory in Ann Arbor, MI.

Richard W. Baldauf

Richard W. Baldauf is an engineer with concurrent positions at the National Vehicle Fuel Emissions Laboratory in Ann Arbor, MI, and the National Risk Management Research Laboratory in Research Triangle Park, NC.