Intermediate Volatility Organic Compound ... - ACS Publications

Aug 31, 2015 - †Center for Atmospheric Particle Studies and ‡Department of Mechanical Engineering, Carnegie Mellon University, 5000 Forbes Avenue,...
1 downloads 11 Views 2MB Size
Article pubs.acs.org/est

Intermediate Volatility Organic Compound Emissions from On-Road Diesel Vehicles: Chemical Composition, Emission Factors, and Estimated Secondary Organic Aerosol Production Yunliang Zhao,†,‡ Ngoc T. Nguyen,†,‡ Albert A. Presto,†,‡ Christopher J. Hennigan,†,‡,§ Andrew A. May,†,‡,∥ and Allen L. Robinson*,†,‡ †

Center for Atmospheric Particle Studies and ‡Department of Mechanical Engineering, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, Pennsylvania 15213, United States S Supporting Information *

ABSTRACT: Emissions of intermediate-volatility organic compounds (IVOCs) from five on-road diesel vehicles and one off-road diesel engine were characterized during dynamometer testing. The testing evaluated the effects of driving cycles, fuel composition and exhaust aftertreatment devices. On average, more than 90% of the IVOC emissions were not identified on a molecular basis, instead appearing as an unresolved complex mixture (UCM) during gaschromatography mass-spectrometry analysis. Fuel-based emissions factors (EFs) of total IVOCs (speciated + unspeciated) depend strongly on aftertreatment technology and driving cycle. Total-IVOC emissions from vehicles equipped with catalyzed diesel particulate filters (DPF) are substantially lower (factor of 7 to 28, depending on driving cycle) than from vehicles without any exhaust aftertreatment. Total-IVOC emissions from creep and idle operations are substantially higher than emissions from high-speed operations. Although the magnitude of the total-IVOC emissions can vary widely, there is little variation in the IVOC composition across the set of tests. The new emissions data are combined with published yield data to investigate secondary organic aerosol (SOA) formation. SOA production from unspeciated IVOCs is estimated using surrogate compounds, which are assigned based on gas-chromatograph retention time and mass spectral signature of the IVOC UCM. IVOCs contribute the vast majority of the SOA formed from exhaust from on-road diesel vehicles. The estimated SOA production is greater than predictions by previous studies and substantially higher than primary organic aerosol. Catalyzed DPFs substantially reduce SOA formation potential of diesel exhaust, except at low speed operations.



INTRODUCTION Intermediate-volatility organic compounds (IVOCs) are species with effective saturation concentrations between 103 and 106 μg/m3, roughly the same range as C12 ∼ C22 n-alkanes.1−3 Laboratory photo-oxidation experiments using both individual compounds4−6 and dilute tailpipe emissions from vehicles and small engines7−9 indicate that IVOCs directly emitted by sources (primary IVOCs) are an important class of secondary organic aerosol (SOA) precursors. Their contribution to atmospheric SOA formation was recently substantiated by direct measurements of IVOC concentrations in the Los Angeles area.3 On-road diesel vehicles are likely an important source of IVOCs in urban areas.10−12 Although many studies have measured emissions of individual IVOCs (e.g., large n-alkanes and polycyclic aromatic hydrocarbons),13−15 few studies have directly measured the total-IVOC emissions,11 mainly due to difficulties in both quantitative collection and chemical analysis of IVOCs. IVOCs cannot be quantitatively collected using techniques commonly employed for measuring VOCs and organic aerosol (OA).16 Furthermore, the majority of the © XXXX American Chemical Society

IVOC emissions cannot be speciated using traditional chromatography based techniques, instead appearing as an unresolved complex mixture (UCM) of coeluting compounds.3,11 Emissions of total (speciated and unspeciated) IVOCs from diesel vehicles (and other sources) have been estimated based on scaling measured emissions of primary organic aerosol (POA)11,17,18 or nonmethane hydrocarbons (NMHCs).10,19 However, without direct measurements of IVOCs, the mass ratio between total IVOCs and either POA or NMHCs is poorly constrained. Emissions from diesel vehicles (including POA and NMHCs) have been reduced substantially in recent years with the implementation of strict new emission standards. To meet these emission standards, new diesel vehicles are equipped with aftertreatment technologies such as catalyzed Received: June 10, 2015 Revised: August 19, 2015 Accepted: August 31, 2015

A

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology diesel particulate filters (DPF).20−22 Although previous studies have investigated the effects of DPFs on emissions of speciated IVOCs,13 their effects on the total-IVOC emissions need to be better understood.3 In this study, the total (speciated and unspeciated) IVOC emissions from both heavy-duty and medium-duty diesel vehicles (HDDVs and MDDVs) were measured during chassis dynamometer testing. The testing examined different driving cycles, fuel compositions and exhaust aftertreatment devices. To evaluate approaches commonly used to develop emission inventories for total IVOCs, the relationship between emissions of total IVOCs with NMHCs, POA, and individual speciated IVOCs was investigated. SOA production from the measured IVOCs was estimated and compared with data from photooxidation experiments performed with dilute tailpipe emissions inside of a smog chamber. Data for traditional emissions (such as POA and NMHCs), gas-particle partitioning, and SOA production from these tests have been reported previously.8,22,23

any aftertreatment device to reduce emissions. The TRU was tested on an engine dynamometer using the 4-mode EPA TRU cycle and a commercial ultralow sulfur diesel fuel comparable to the midaromatic diesel fuel.22 Emission Characterization. The entire exhaust from each vehicle was diluted in a constant volume sampling (CVS) system with air treated by high-efficiency particulate air filters following Code of Federal Regulations Title 40, Chapter 1, Subchapter C, Part 86.25 The dilution ratio varied in time and by cycle; the test-average ratio was roughly 30, 8, 16, and 100 to 1 during UC, UDDS, cruise and creep+idle cycles, respectively.22 Detailed descriptions of the experimental setup, the operation of the chassis dynamometers, and comprehensive emission measurements have been published elsewhere.8,22 This study focuses on previously unreported IVOC emissions data. IVOCs were sampled by drawing dilute exhaust from the CVS through a quartz filter immediately followed by two adsorbent tubes (Gerstel 6 mm OD, 4.5 mm ID glass tube filled with ∼290 mg of Tenax TA), all connected in series. The samples were collected over the entire drive cycle and therefore represent trip-average emissions. Two identical sampling trains were employed to collect duplicate samples. Both sampling trains (including the transfer line from the CVS) were electrically heated to ∼47 ± 5 °C to match the CFR 86 protocol.26 After sampling, the adsorbent tubes and quartz filters were stored at −18 °C. The second adsorbent tube was used to determine breakthrough of IVOCs during sampling. The IVOC mass measured on the backup tube averaged 6 ± 3% of total collected IVOCs for nonaftertreatment vehicles and 33 ± 12% for DPF vehicles. The IVOC emissions reported here are the sum of the emissions measured on both adsorbent tubes. Laboratory blanks were analyzed to verify that there was negligible residual carbon on both the prefired quartz filters and adsorbent tubes prior to sampling. Dynamic blanks (both filters and adsorbent tubes) were collected using the same sampling system while operating the CVS just on dilution air for the same period of time as a standard test cycle. In addition to the IVOC data, our analysis used NMHCs and CO2 emission data reported in May et al.22 and POA emission data from the quartz filter located upstream of the adsorbent tubes. NMHCs were defined as the difference between total hydrocarbons and CH4. Total hydrocarbons were measured by heated flame ionization detection (FID), CH4 by gas chromatography-FID, and CO2 by nondispersive Infrared (NDIR) detectors (IRD-4000).22 POA emissions were determined by analyzing the quartz filters using a Sunset Laboratory OC/EC Carbon Aerosol Analyzer Model 3 and the IMPROVE-A protocol.27 POA was estimated by multiplying the measured organic carbon by an organic-mass-to-carbon ratio of 1.2.23 Quantification of IVOCs. Adsorbent tubes were analyzed by a gas chromatography/mass spectrometer (GC/MS) (Agilent, 6890 GC/5975 MS) equipped with a thermal extraction and injection system (Gerstel, Baltimore, MD) and a capillary column (Agilent HP-5MS, 30 m × 0.25 mm) following an established protocol.3 Briefly, a known amount of deuterated standards (d8-naphthalene and C12, C16, C20, C24, C30, C32, C36 deuterated n-alkanes) was injected into each adsorbent tube prior to thermal desorption to determine the recovery of IVOCs during analysis. The adsorbent tube sample was thermally desorbed at 275 °C inside a Gerstel thermal



MATERIALS AND METHODS Test Fleet, Fuels and Test Cycles. Measurements of primary IVOC emissions were conducted for both HDDVs and MDDVs during chassis dynamometer testing at the California Air Resources Board (CARB) Heavy-duty Engine Testing and Haagen-Smit Laboratories. Detailed information for the test fleet, fuels, test cycles can be found elsewhere.8,22 A brief description is provided here. The test fleet consisted of three HDDVs and two MDDVs (Table S1). Although this is a small test fleet, the vehicles were selected to span a range of emission control technologies. Three diesel vehicles were equipped with exhaust aftertreatment devices. The 2010 HDDV was equipped with an original equipment manufacturer (OEM) installed catalyzed diesel particulate filter (DPF) with a selective catalytic reduction (SCR) system. The 2007 HDDV had an OEM installed catalyzed DPF. The 2005 MDDV was equipped with a diesel oxidation catalyst (DOC). One HDDV and one MDDV had no aftertreatment devices. The DOC on the 2005 MDDV appeared to be compromised;8 therefore, for discussion, it is grouped with the two no aftertreatment vehicles into a category referred to as nonaftertreatment vehicles. Three types of ultralow sulfur diesel fuels, differentiated by their aromatic content, were used during tests of HDDVs: lowaromatic (9% aromatic content), midaromatic (12% aromatic content) and high-aromatic (28% aromatic content) diesel fuel.22 A commercial ultralow sulfur fuel purchased from a local gas station was used for tests of MDDVs, which was comparable, but not identical, to the midaromatic diesel fuel. HDDVs were tested using the hot-start Urban Dynamometer Driving Schedule (UDDS), which simulates typical urban driving. HDDVs were also tested using select modes of the Heavy Heavy-Duty Diesel Truck (HHDDT) driving schedule to investigate IVOC emissions during low-speed (creep+idle) and high-speed (high-cruise) operations. MDDVs were tested over a single cold-start unified cycle (UC), which represents the typical urban driving patterns in southern California. For discussion, we group results from UDDS, UC, and high-cruise cycles together, referring to them collectively as higher-speed cycles. The driving schedules used in this study are listed in Table S2. An off-road diesel engine for a transportation refrigeration unit (TRU) was also tested.24 The TRU was not equipped with B

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

retention time.3 Following quantification of the total-IVOC mass, the UCM mass is estimated by subtracting quantified speciated IVOCs from total-IVOC mass in each bin. The volatility distribution of the IVOCs is derived using the effective saturation concentration (C*) of the n-alkane in each bin as a surrogate for the volatility of the IVOCs in that bin (Figure 2).2,3

desorption unit (TDS3) and then preconcentrated by a Gerstel cooled injection system (CIS4) at −120 °C downstream of the TDS3, followed by thermal injection into GC/MS for analysis. Figure 1 shows a typical chromatogram of an adsorbent tube sample. The peaks correspond to individual compounds. The

Figure 2. Mass fraction distributions of organics thermally desorbed during GC/MS analysis from the quartz filter/adsorbent tube sample sets from all of the nonaftertreatment diesel tests. The main figure shows the distributions based on elution time (Table S4); inset shows the distributions plotted in the volatility basis set framework (Table S5). The boxes represent the 75th and 25th percentiles with the centerline being the median. The whiskers are the 90th and 10th percentiles. The gray-shaded area represents the median mass fraction of organics desorbed from quartz filters. The red-shaded area indicates the median SVOC breakthrough from the quartz filters. Figure 1. (a) Chromatogram of total ion current; (b) quantified IVOC components in each retention-time bin; and (c) average mass spectrum of an adsorbent sample from a nonaftertreatment diesel vehicle (model year 2006) operated over the UDDS. The vertical dash lines in (a) mark the width of the retention time bins plotted in (b).

The IVOC UCM within each retention-time bin is further separated into two chemical groups based on the mass spectrum: the unspeciated b-alkanes and remaining UCM (Figure 1b). The remaining UCM is presumably dominated by cyclic alkanes, 28 but likely also contains other cyclic compounds, such as alkyl benzenes. We therefore refer to the remaining UCM as “unspeciated cyclic compounds”. The mass of unspeciated b-alkanes is estimated using the signal of the C4H9+ fragment (m/z 57) in each UCM bin minus the m/z 57 signal from all identified n- and b-alkanes in that same bin and the average fraction of m/z 57 in the mass spectra in a suite of b-alkanes standards.3 This approach assumes that the entire m/ z 57 signal is due to n- and b-alkanes. It likely is an upper bound estimate of the unspeciated b-alkane mass because compounds other than n- and b-alkanes also produce m/z 57 fragments. Unspeciated cyclic compounds are defined as the difference between the total UCM and unspeciated b-alkanes in each bin. Quartz filter samples were also analyzed via GC/MS using the same method as IVOCs except that the filters were thermally desorbed at 300 °C instead of 275 °C. By combining the GC/MS data from the quartz filters and adsorbent tubes, we can construct a volatility distribution of IVOCs, semivolatile organic compounds (SVOCs) and low-volatility organic compounds (LVOCs) (Figure 2). Emission Factor (EF). The IVOC data are reported as fuelbased emission factors (EF, mg/kg-fuel) calculated using the carbon-mass-balance approach:

resolved compounds included straight chain (n-alkanes) and branched alkanes (b-alkanes), alkylcyclohexanes, unsubstituted and substituted polycyclic aromatic hydrocarbons (PAHs), and alkylbenzenes (Supporting Information). Emissions of individual IVOCs were quantified based on the calibrated instrument response to authentic standards and accounting for recovery of the deuterated internal standard. However, the sum of speciated IVOCs accounts for less than 10%, on average, of the total IVOC signal (Figure 1a, b). The predominant fraction of total IVOC signal appears as an unresolved complex mixture (UCM) (Figure 1a), similar to ambient IVOC samples collected during CalNex.3 The total (speciated + unspeciated) IVOC mass was quantified using the approach of Zhao et al.3 Briefly, the total ion signal is integrated into bins based on retention time acquired during the GC analysis.3,28 Eleven retention-timebased bins are defined, each one corresponding to an n-alkane (C12-C22) (Figure 1a). The IVOC bin corresponding to the Cn n-alkane (the subscript “n” denotes the number of carbon) is designated as the Bn bin (Figure 1a). The total IVOC mass in the Bn bin is calculated using the mass/signal response of the Cn n-alkane.3 This approach accounts for the variation in the mass/ signal response of hydrocarbons in the GC/MS as a function of C

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology EFIVOCs =

emission data for individual tests are compiled in Table S3 and S4. The DOC-equipped MDDV is included in the nonaftertreatment category because the DOC was compromised.8 The error bars in Figure 3a indicate that, within each of these four categories, there was little variation in the magnitude of the IVOC. The one exception was creep + idle operations of the two DPF-equipped vehicles (Table S4). The two vehicles equipped with catalyzed DPFs have dramatically lower total-IVOC emissions than all three vehicles without aftertreatment devices. For example, the average total IVOC EFs for the nonaftertreatment vehicles operated over the higher speed cycles (UDDS, UC and High-cruise) is about 28 times greater than the DPF vehicles (696 ± 102 versus 25 ± 8 mg/kg-fuel) (Figure 3a). This difference is much smaller (a factor of 7) if the vehicles are operated over the creep+idle cycle (3933 ± 1000 versus 586 (two tests) mg/kg-fuel). The smaller difference at creep+idle is likely due to reduced DPFcatalyst performance at low exhaust temperatures. However, despite reduced catalyst performance at creep+idle, the data demonstrate that the catalyzed DPF significantly reduced the IVOC emissions across the entire range of driving conditions considered here. The DOC-equipped MDDV had similar IVOC emissions as vehicles without exhaust aftertreatment (Table S4). This was unexpected as previous studies have shown that DOCs substantially reduce hydrocarbon emissions.30,31 We therefore expected a reduction in IVOC emissions similar to that of the catalyzed-DPF vehicles. The DOC on this vehicle was thought to be compromised,8 highlighting that vehicles with faulty aftertreatment devices (DOC or DPF) likely have IVOC emissions similar to vehicles without exhaust aftertreatment. Figure 3a also shows that IVOC emissions during creep+idle operations are much higher than emissions during higher speed operations. For example, the average IVOC emissions during the creep+idle cycle are a factor of ∼6 and 23 higher than other driving cycles for nonaftertreatment and DPF-equipped vehicles, respectively. These differences are important as roughly 7% of on-road diesel fuel is consumed during vehicle idling and low-speed driving.32 Approximately 30% of the totalIVOC emissions from on-road diesel vehicles occur during lowspeed operations. This percentage will increase as the fraction of DPF diesel vehicles in the on-road fleet increases. Although the DPF-equipped vehicles had very low IVOC emissions at high speed operations, their IVOC emissions were substantially higher during creep+idle operations; on average, comparable to those from nonaftertreatment vehicles operated over higher-speed cycles. This indicates that the DPF-catalyst did not efficiently oxidize IVOCs at low exhaust temperatures associated with low speed operations. However, the creep+idle IVOC emissions from the DPF-equipped vehicles were highly variable (105 versus 1067 mg kg-fuel−1; Table S4), indicating significant vehicle-to-vehicle differences in catalyst performance during low-speed operation. Total Organic Emissions. IVOCs are a major component of the total organic emissions from diesel vehicles (Table S4). The average (±one standard deviation) ratio of IVOCs to NMHCs and POA is 0.6 ± 0.1 and 12 ± 7, respectively, for the nonaftertreatment diesel vehicles, and 1.5 ± 0.8 and 31 for DPF-equipped diesel vehicles. For the DPF-equipped vehicles, the average IVOC-to-NMHC ratio is greater than one; this could be caused by background IVOCs inside the CVS, which corresponds up to 42% of measured IVOCs during the UDDS. However, the average IVOC-to-NMHC ratio could also be due

[IVOCs] f [ΔCO2 ] c

where [IVOCs] is the measured mass concentration of IVOCs in the CVS; [ΔCO2] is the measured background-corrected CO2 concentration in the CVS expressed in units of carbon mass (CO2 accounts for over 99% of the carbon emissions in these experiments); and fc is the mass fraction of carbon in the diesel fuel (0.85).22 Measured fuel economy from each test is reported in Table S2 to convert between fuel- and distancebased EFs. The IVOC data were not corrected for the background signal measured on the dynamic blanks (Supporting Information). These blanks are small for the nonaftertreatment equipped vehicles (less than 5% of IVOC signal) but can be significant for the tests with DPF-equipped vehicles (up to 42% of the IVOC during higher speed tests). Therefore, the data may overestimate the IVOC emissions, especially for DPF-equipped vehicles. However, dynamic blanks measured while operating the CVS on clean air likely overestimate background concentrations.29



RESULTS Figure 3a compares the measured total IVOC (speciated + unspeciated) emission factors. The results are presented as averages across all tests grouped by vehicle type (DPF versus nonaftertreatment) and driving cycle (creep + idle versus the higher-speed cycles − UDDS, UC, and high-speed cruise). The

Figure 3. (a) Average emission factors of total (speciated + unspeciated) IVOCs; across all tests grouped by the vehicle type (DPF versus nonaftertreatment) and drive cycle (creep + idle versus the higher-speed cycles − UDDS, UC and high-speed cruise); (b) average composition of the total-IVOC emissions and (c) estimated relative contribution of DPF-equipped diesel vehicles to total IVOCs emitted from on-road diesel vehicles and reduction in total-IVOC emissions from diesel vehicles as a function of diesel fuel consumption. Error bars in (a) and (b) are one standard deviation of measured data. The upper and lower bounds of the filled areas in (c) represent the range of contributions of DPF vehicles based on the variability of the measured DPF-vehicle creep + idle emissions. D

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

matter emissions. Quantitative SVOC emissions data can be used to predict atmospheric POA concentrations through the partitioning theory.33 Figure 2 also shows that quartz filters collect a small fraction of the IVOC emissions, less than 3% of total IVOCs (quartz filter + adsorbent tubes) for both nonaftertreatment and DPFequipped vehicles. These are presumably adsorbed vapors (positive sampling artifact).34 Quartz filters also collect LVOCs (>B32), which were mainly measured in the nonaftertreatment vehicle exhaust (Figure 2, Figure S2). To quantify the recovery of LVOCs in the GC/MS analysis, the amount of organics measured by the GC/MS from quartz filter samples was compared to data from EC/OC analysis of the same samples. For the nonaftertreatment vehicles, the GC/MS analysis recovered an average of 57% ± 29% of organics measured by the EC/OC analyzer from the same filter sample, consistent with similar analysis of POA in a traffic tunnel.35 Given that the GC/MS analysis uses a lower thermal desorption temperature than the EC/OC analysis (300 °C versus 580 °C), organics not recovered during the GC/MS analysis are likely LVOCs. This hypothesis is consistent with the thermodenuder data for these vehicles.23 Although the unrecovered material represents about 4% of the total organics with a C* less than 3 × 10−6 μg m−3, it is an important component of POA. Therefore, it is important to analyze quartz filters using both thermal desorption GC/MS and an EC/OC analyzer to better predict POA emissions. Chemical Composition of IVOCs. The individual species identified from the GC/MS analysis contributed, on average, 5.6% ± 3.1% and 8.1% ± 2.3% of the total IVOC mass emissions from DPF-equipped and nonaftertreatment diesel vehicles, respectively (Table S3). The speciated IVOCs are mainly n-alkanes, b-alkanes, PAHs and substituted PAHs. Naphthalene and substituted naphthalenes account for about 90% of quantified PAHs and substituted PAHs (Table S3). There is a systemic trend in IVOC composition with fuel composition, but no statistically significant trends with drive cycle. For example, the mass fraction of PAHs to total IVOC correlates to the fuel aromatic content; the average fractions of the sum of individual speciated PAHs and substituted PAHs from nonaftertreatment diesel vehicles are 0.8% ± 0.1%, 1.4% ± 0.2% and 5.1% ± 0.4% for low-, mid- and high-aromatic diesel fuels. For DPF-equipped vehicles, the average fraction of speciated PAHs and substituted PAHs is 0.4% (one test), 0.5% ± 0.2%, and 0.9% ± 0.4% for low-, mid- and high-aromatic diesel fuels. The fractions of speciated PAHs and substituted PAHs to total IVOCs for DPF-equipped vehicles are less than those for nonaftertreatment ones, but the relative abundance of these compounds, such as naphthalenes, shows a good correlation between DPF-equipped and nonaftertreatment diesel vehicles. More than 90% of the IVOC emissions could not be speciated using the traditional GC/MS technique, instead appearing as an unresolved complex mixture (UCM). The chemical characteristics of the IVOC UCM are illustrated by the average mass spectrum of the total IVOCs plotted in Figure 1c. This mass spectrum has very strong CnH2n+1, CnH2+n−1 and CnH2n−3 sequences and m/z 57 is the most abundant mass fragment; these are characteristics of electron ionization of aliphatic compounds.36 Therefore, the majority of the IVOC UCM is likely branched and cyclic alkanes, consistent with diesel fuel composition.8 All of the adsorbent samples have mass spectra similar to Figure 1c (Figure S3).

to only a portion of the IVOCs being measured as NMHCs. The much higher IVOC-to-POA ratio for DPF-equipped vehicle underscores the high efficiency at which the catalyzed DPF removes POA. The GC/MS data from the adsorbent tubes and quartz filters can be combined to quantify the emissions of all organics with C* less than 3 × 10−6 μg m−3 (B12). This includes three classes of organics based on volatility: IVOCs (C* = 300−3 × 106 μg m−3); SVOCs (C* = 0.3−300 μg m−3); and LVOCs (C* < 0.3 μg m−3) (Figure 2). Figure 2 presents a box-whisker plot of the volatility distribution of organics desorbed from quartz filters and adsorbent tubes from all nonaftertreatment vehicle tests (Tables S4 and S5). Although the magnitude of the IVOC emissions varies widely (Figure 3a), the volatility distribution for nonaftertreatment vehicles is well constrained (relative standard deviations are less than 25% across the entire set of tests for bins with more than 9% of total-IVOC emissions). The IVOCs account for the vast majority (93 ± 1%) of the total organic emissions with a C* less than 3 × 10−6 μg m−3 (IVOCs + SVOCs + LVOCs) followed by SVOCs (6.2 ± 1.3%) and LVOCs (0.8 ± 0.3%). The fact that IVOC emissions dominate SVOC/LVOC emissions indicates that SOA production will ultimately exceed the contribution of the direct POA emissions to ambient fine particulate matter. DPF-equipped vehicles have a similar IVOC distribution as nonaftertreatment equipped vehicles (Figure S1; Tables S4 and S5), except that the DPF-vehicle data are more variable because of the lower emission rate (lower signal-to-noise). The DPFvehicles do show a shift in the IVOC volatility distribution with drive cycle not observed with nonaftertreatment vehicles; for example, IVOCs in B12-B14 IVOC bins account for 70%, on average, of the total-IVOC emissions during the creep+idle cycle versus only 41% of total IVOCs during the UDDS cycle (Table S4). This shift is presumably due to changes in catalyst efficiency with changing exhaust temperature. Figure S2 shows the volatility distribution of IVOCs, SVOCs, and LVOCs for DPF-equipped vehicles with corresponding relative contributions to total organics as 74 ± 21%, 26 ± 21%, and 0.1 ± 0.1%. The enhanced mass fraction of SVOCs relative to nonaftertreatment vehicles suggests the different removal efficiency of the catalyst-DPF for different organics. Quartz filters are commonly used to characterize POA emissions, which are comprised of SVOCs and LVOCs.33 However, Figure 2 shows that quartz filters collected at ∼47 °C do not capture all of the SVOC emissions (B23 ∼ B32). In these experiments, the quartz filters collected, on average, only 48% ± 7% of the total SVOCs (filter + adsorbent tubes) emissions from nonaftertreatment diesel vehicles and 11% ± 9% during tests of DPF-equipped vehicles. The low collection efficiency is presumably due to some of SVOCs existing as vapors at 47 °C inside the sampling system. The difference in the SVOC collection efficiency between the DPF and nonaftertreatment equipped vehicles is likely due to differences in POA and black carbon emissions (the much lower primary particulate matter emissions from DPF-vehicles reduce adsorption of SVOCs). The low capture of SVOCs by quartz filters highlights a potential weakness of existing particulate matter emission certification protocols (e.g., CFR 86)26 that require filter collection at 47 °C (mainly to prevent water condensation). If filters are collected at 47 °C, one must also collect downstream adsorbent tubes to quantitatively capture the SVOC emissions, especially for DPF vehicles with very low primary particulate E

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

importance of off-road diesel engines as a source of IVOCs will likely increase. Off-road diesel engines are rarely equipped with DPFs. Estimating Total-IVOC Emissions Using POA, NMHCs, and Speciated IVOC Data. Although IVOCs are an important class of SOA precursors,3,10 very limited data exists for total (speciated + unspeciated) IVOC emissions. Consequently, previous SOA modeling studies have estimated totalIVOC emissions by scaling measured POA, NMHCs, or individual IVOCs.10,18,39 However, these estimates are poorly constrained due to lack of direct measurements of total IVOCs in tailpipe emissions. The simultaneous measurements of total IVOCs, POA, and NMHCs performed here allow evaluation of the different scaling approaches. The following sections focus on estimating total-IVOC emissions from nonaftertreatment diesel vehicles because nonaftertreatment diesel vehicles dominate IVOC emissions from on-road diesel vehicles (Figure 3c). Relationship between Total IVOCs and POA. Robinson et al.17 estimated total-IVOC emissions by scaling POA emissions. Figure 4a shows that there is some correlation between total

To further characterize the composition of the IVOC UCM, these emissions were classified into 22 groups: unspeciated balkanes and unspeciated cyclic compounds in each of the 11 retention-time-based bins (Figure 1b). More than 75% of the total-IVOC emissions (speciated + unspeciated) are classified as unspeciated cyclic compounds (Figure 3b). Therefore, the IVOC UCM measured in every test is dominated by unspeciated cyclic compounds (Figure 3b). Unspeciated balkanes are estimated to contribute less than 20% of the emissions. There is some dependence on the composition of the IVOC UCM on fuel composition. For example, unspeciated cyclic compounds contribute 81.3% ± 0.7%, 76.4% ± 1.6%, and 77.8% ± 0.7% of tailpipe IVOCs from nonaftertreatment vehicles operated with low-, mid- and high-aromatic diesel fuels. A similar trend (88.0% (one test), 79.9% ± 10.0%, 74.7% ± 6.4%) is observed for DPF-diesel vehicles operated on low-, mid- and high-aromatic diesel fuels. Although the chemical composition of IVOC emissions shows some dependence on fuel composition, the GC/MS data reveal significant differences in both molecular and bulk IVOC composition between exhaust and diesel fuel (Supporting Information). This underscores the importance of characterizing the chemical composition of the actual emissions for estimating SOA production from diesel vehicles.



DISCUSSION Reducing IVOC Emissions. Since the total-IVOC emissions for DPF-equipped diesel vehicles are dramatically lower than nonaftertreatment diesel vehicles, the recent regulations that effectively require equipping on-road vehicles with DPFs should reduce total-IVOC emissions from on-road diesel vehicle fleet. Figure 3c shows the predicted reduction in total-IVOC emissions as a function of the fraction of diesel fuel consumption in DPF-equipped vehicles. The predictions are based on the average total IVOC EFs for DPF and nonaftertreatment vehicles shown in Figure 3a. The shaded area indicates the full range of estimates based on data from individual tests of DPF-equipped diesel vehicles over the creep +idle cycle. Seven percent of the total diesel consumption is assumed to occur during idling and low-speed driving.32 As expected, Figure 3c indicates that shifting the on-road fleet toward DPF vehicles reduces the IVOC emissions relative to a fleet solely consisting of nonaftertreatment diesel vehicles. However, on a relative basis, the total-IVOC emissions from on-road diesel vehicles will be dominated by nonaftertreatment diesel vehicles even if DPF-equipped diesel vehicles consume 90% of on-road diesel fuel. Figure 3c focuses on on-road vehicles, but long-term trends reveal the increasing importance of off-road engines.37 This likely also applies to IVOC emissions. Off-road diesel engines consumed 36% of the distillate fuels in the United States in 2013.38 The TRU engine has the same total-IVOC emissions as the nonaftertreatment vehicles (Figure 3a). Assuming that IVOC emissions from the TRU are representative of all offroad engines, off-road diesel sources would likely dominate total-IVOC emissions from diesel engines in the future. The most recent EPA off-road exhaust standards (TIER 4) limit NMHC emission standards from larger diesel off-road engines, but not for smaller off-road diesel engines ( 0.8). However, the slopes of the regression varied with the chemical composition of diesel fuel (Table S6). Therefore, the relative consumption of different types of diesel fuel should be accounted if total-IVOC emissions are estimated using speciated IVOC data. Estimating Emissions from Off-Road Engines. Although the total-IVOC EF for the TRU is similar to the nonaftertreatment vehicles (Figure 3a), the ratios of IVOCs to POA and NMHCs for the TRU are 2.7 and 0.25 versus 9.5 and 0.6 for the nonaftertreatment vehicles. This suggests that there may be important differences between off-road and on-road engines, which likely reflect differences in emission standards. Off-road sources have been subject to less stringent emission standards than on-road vehicles. This hypothesis is supported by the fact that the POA and NMHC emissions from this TRU (model year 1998) are more similar to older on-road diesel engines tested Schauer et al.11 than the newer nonaftertreatment onroad engines tested here. More IVOC emissions data are clearly needed for off-road engines, but the TRU measurements G

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

IVOC mass; Table S4), the SOA production largely depends on the magnitude of the IVOC emissions and the contributions of the different precursors categories are essentially the same as that shown in Figure 5b. Therefore, SOA production during the creep+idle cycle is higher than other cycles (Figure S4) because of higher IVOC emissions (Figure 3a). The model predicts that DPF-equipped vehicle have essentially no SOA production at higher speed operations (consistent with the smog chamber experiments of Gordon et al.8) (Figure 5c). The model predicts substantial SOA formation from DPF-equipped vehicle exhaust from lowspeed operations, but still less than exhaust of nonaftertreatment vehicles undergoing higher speed operations. We are not aware of any SOA measurements for idling DPF-equipped vehicles. The model indicates that SOA production in the atmosphere likely exceeds direct POA emissions. For example, after all of the IVOCs are oxidized the SOA-to-POA ratios are 3.2 ± 1.7 and 7.6 ± 1.2 for nonaftertreatment vehicles undergoing higher-speed and creep+idle operations, respectively. For DPFequipped vehicles, the SOA-to-POA ratios are even larger. The average SOA-to-POA ratio after all of IVOCs are oxidized is 5.0 and 31 during the UDDS and creep+idle cycle, respectively. Therefore, SOA will become (relatively) even more important in the future as DPF-equipped vehicles become more common. In addition, the SOA-to-POA ratios reported above likely underestimate the true importance of SOA production because the model does not account for the semivolatile nature and partitioning biases of POA17,23 and multigenerational oxidation on SOA formation.43 Figure 5a also compares the predicted SOA formation to data from smog chamber experiments of Gordon et al.8 who photooxidized dilute tailpipe emissions from the nonaftertreatment vehicles operated over UDDS, UC, and high cruise driving cycles. The predicted SOA production from IVOCs is substantially greater than the measured SOA. The average ratio between predicted and measured SOA is 1.8 ± 0.6 for nonaftertreatment vehicles operated over the higher-speed driving cycles. The difference was even larger for nonaftertreatment vehicle creep+idle experiments (model-to-measurement ratio of 2.3). There are several possible explanations for this modelmeasurement discrepancy. First, the smog chamber experiments of Gordon et al.8 may underestimate the SOA production from actual diesel exhaust. Although Gordon et al.8 corrected the smog chamber data for both vapor and particle wall losses, vapor losses to Teflon chamber walls are uncertain.44 In addition, the dilute exhaust was transferred from the CVS (where the IVOC emissions were measured) into the smog chamber (where the SOA formation occurred) through a heated (47 °C) and passivated sampling line.8 The model calculations assume no losses in this transfer line; however, transfer line losses would reduce SOA precursor concentrations inside the chamber, which, in turn, would reduce the measured SOA formation. The loss of IVOCs in the transfer line likely depends on compound volatility, with greater losses expected for lower-volatility compounds, such as IVOCs.45,46 Another potential explanation for the model-measurement discrepancy shown in Figure 5a is that the approach of Zhao et al.3 overpredicts the SOA formation from unspeciated IVOCs. However, Zhao et al.3 argues that the yields for the unspeciated IVOCs are likely conservative (lower bound estimates). For example, the UCM classification approach overestimates the

Figure 5. (a) Comparison of measured and predicted SOA production as a function of photo-oxidation time (calculated using an OH concentration of 1.5 × 106 moles cm−3). (b) Comparison of the average POA emissions and predicted SOA production after 48 h of photo-oxidation. (c) Average SOA production, grouped by vehicle type and drive cycle. Model predictions in (a) and (b) are for average nonaftertreatment diesel vehicles operated over higher speed cycles (UDDS, UC, high-speed cruise). The data in (a) are from smog chamber experiments with nonaftertreatment vehicles operated over UDDS, UC and High-Cruise cycles.8 The gray area in (a) represents the 25th ∼75th percentile range of estimated SOA production based on range of measured precursor EFs and composition. The error bars represents one standard deviation of measured data.

multigeneration chemistry will likely cause continued SOA production, but this chemistry is highly uncertain and therefore not accounted for in the model of Zhao et al.3 Figure 5b presents the average contribution of different classes of precursors to the predicted SOA production for nonaftertreatment vehicles after 48 h oxidation. The vast majority of predicted SOA mass is contributed by IVOCs with only about 1% from single-ring aromatic compounds (C6-C9). Speciated IVOCs contributed ∼7% of the predicted SOA. Unspeciated cyclic compounds dominate the predicted SOA production. Figure 5c compares the predicted SOA production for DPFand nonaftertreatment vehicles undergoing creep+idle and higher speed operations (the same four groups as shown in Figure 3a). The results are the SOA production at high OH exposure (e.g., asymptote in Figure 5a). Time series of SOA production for each group of tests are shown in Figure S4. Given that the mass fractions of unspeciated cyclic compounds are similar across all of the tests (relative standard deviations are less than 40% for bins that contribute more than 5% of H

DOI: 10.1021/acs.est.5b02841 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology Present Addresses

amount of unspeciated b-alkanes, which have lower SOA yields relative to cyclic compounds with the same carbon number.5,41 In addition, some cyclic alkanes may have SOA yields that are greater than n-alkanes with the same carbon number.41,42 Figure S5 compares the new estimate of SOA production to two previous estimates. Jathar et al.19 estimated SOA production from unspeciated NMHC in tailpipe emissions from on-road diesel vehicles using the measured SOA data shown in Figure 5a. Gentner et al.10 estimated SOA production using unburned diesel fuel as the surrogate for tailpipe emissions. The SOA production from diesel exhaust estimated here is about a factor of 2 higher than both previous estimates (Figure S5). Although the magnitude of the differences is similar, the reason is different for each study. Jathar et al.19 used similar SOA yields as the Zhao et al.3 model, but assumed that only 25% of NMHCs are IVOC-like species (unspeciated SOA precursors). The new direct measurements of IVOC emissions reported here indicate that IVOCs correspond to 60% of NMHCs (Figure 4b), more than twice as much assumed by Jathar et al.19 Gentner et al.10 estimated similar amount of IVOC emissions as measured here. However, the yields applied by Gentner et al.10 are lower than the Zhao et al.3 model. First, Gentner et al.10 used yields from chamber experiments performed at very high organic aerosol concentrations (>1000 μg/m3),5 which likely underestimate SOA production.6 Zhao et al.3 used the higher yield data of Presto et al.6 that were measured at more atmospherically relevant OA concentrations (