Emission Factors for High-Emitting Vehicles Based on On-Road Measurements of Individual Vehicle Exhaust with a Mobile Measurement Platform

ABSTRACT

Fuel-based emission factors for 143 light-duty gasoline vehicles (LDGVs) and 93 heavy-duty diesel trucks (HDDTs) were measured in Wilmington, CA using a zero-emission mobile measurement platform (MMP). The frequency distributions of emission factors of carbon monoxide (CO), nitrogen oxides (NO_x), and particle mass with aerodynamic diameter below 2.5 μm (PM_2.5) varied widely, whereas the average of the individual vehicle emission factors were comparable to those reported in previous tunnel and remote sensing studies as well as the predictions by Emission Factors (EMFAC) 2007 mobile source emission model for Los Angeles County. Variation in emissions due to different driving modes (idle, low- and high-speed acceleration, low- and high-speed cruise) was found to be relatively small in comparison to intervehicle variability and did not appear to interfere with the identification of high emitters, defined as the vehicles whose emissions were more than 5 times the fleet-average values. Using this definition, approximately 5% of the LDGVs and HDDTs measured were high emitters. Among the 143 LDGVs, the average emission factors of NO_x, black carbon (BC), PM_2.5, and ultrafine particle (UFP) would be reduced by 34%, 39%, 44%, and 31%, respectively, by removing the highest 5% of emitting vehicles, whereas CO emission factor would be reduced by 50%. The emission distributions of the 93 HDDTs measured were even more skewed: approximately half of the NO_x and CO fleet-average emission factors and more than 60% of PM_2.5, UFP, and BC fleet-average emission factors would be reduced by eliminating the highest-emitting 5% HDDTs. Furthermore, high emissions of BC, PM_2.5, and NO_x tended to cluster among the same vehicles.

IMPLICATIONS

This study presents the characterization of on-road vehicle emissions in Wilmington, CA, by sampling individual vehicle plumes. Approximately 5% of the vehicles were high emitters, whose emissions were more than 5 times the fleet-average values. These high emitters were responsible for 30% and more than 50% of the average emission factors of LDGVs and HDDVs, respectively. It is likely that as the overall fleet becomes cleaner due to more stringent regulations, a small fraction of the fleet may contribute a growing and disproportionate share of the overall emissions. Therefore, long-term changes in on-road emissions need to be monitored.

INTRODUCTION

On-road motor vehicles are a major source of air pollution both locally and nationally. They are responsible for approximately half of carbon monoxide (CO) and nitrogen oxides (NO_x), and 9% of particle mass with aerodynamic diameter below 2.5 μm (PM_2.5) emitted in California.1 Recent studies revealed human exposure to traffic-related air pollutants is associated with a range of health outcomes, including adverse respiratory effects2,3 and increased risk of cardiopulmonary diseases.4 In addition, carbon dioxide (CO₂), NO_x, PM_2.5, and black carbon (BC) emitted by motor vehicles contribute to climate change both directly and indirectly.5

Although overall fleet-average vehicle emissions have declined significantly due to increasingly strict emission standards and improved control technologies, several researchers have suggested that a small, malfunctioning fraction of the fleet contributes a disproportionate fraction of overall vehicle emissions.6–11 Emission factor data from recent remote sensing studies show that the skewness of emission distributions for CO, NO_x, and hydrocarbons has increased over the last decade due to high-emitting vehicles, whereas fleet-average emissions have decreased considerably.10 However, relatively few measurements have been made of the emission characteristics of high emitters under a full range of in-use driving conditions.

Various approaches have been used to measure vehicular emissions. Chassis dynamometer experiments can evaluate emission control technologies over different driving conditions and cycles while providing the advantage of precise and controlled testing conditions.12–14 However, due to their complexity and high cost, these tests are not well suited for capturing a large sample of on-road vehicles and to characterize the variation in fleet emissions. In contrast, tunnel studies can measure a large sample of on-road vehicles, but they are typically limited to specific driving conditions and provide mostly fleet-average results.9,15,16 Remote sensing technology has the ability to screen a large number of individual vehicles quickly, but is generally restricted to measuring only one location at a time and therefore usually only a single operating condition.10,17 To overcome some of these obstacles, researchers have turned to a mobile on-road approach to obtain realistic fleet emission estimates of individual vehicles under many different in-use operating conditions.18–22 However, previous mobile laboratory studies have been limited to measuring relatively small samples of individual vehicle emissions for only specific pollutants such as PM_2.5 and ultrafine particles (UFPs), defined as particles with aerodynamic diameter below 0.1 μm. Also, these “follower” mobile units, if operating with combustion engines, can compromise the results due to self-contamination from their own exhaust, especially when they are idling.

Recently, a zero-emission electric vehicle outfitted with real-time instruments has been used to investigate air pollutant concentrations on roadways and determine human exposure at the individual and subcommunity levels near busy roadways.23–26 In the current study, we utilized a similar zero-emission mobile measurement platform (MMP) to calculate fuel-based emission factors from a large number of individual vehicles under a wide range of in-use driving conditions. CO, NO_x, BC, PM_2.5, and UFP emission factors were measured from 143 light-duty gasoline-powered vehicles (LDGVs) and 93 heavy-duty diesel trucks (HDDTs), in the Wilmington area of Los Angeles, CA, using the zero-emission MMP outfitted with fast-time-response analytical instrumentation. We then compared the resulting fleet-average emission factors to previous studies, calculated the contributions of high emitters to the overall average emission factors, and examined the relationships between high BC, NO_x, and PM_2.5 emitting vehicles.

METHODS

Mobile Measurement Platform

A zero-emission electric vehicle (2002 Toyota RAV4 sub-sports utility vehicle) was used as the mobile platform to eliminate the possibility of the platform's own emissions affecting the measurements. Instruments were chosen for compact size, high time resolution, proven robust mobile operation, and low power consumption. Table 1 describes the instruments and equipment deployed on the mobile platform.

Table 1. Instruments on the mobile measurement platform

Instrument	Measurement Parameter	Instrument Flow Rate (liters per minute)	Response Time (seconds)	Resolution	Detection Limit
TSI portable CPC (butanol-based), model 3007	UFP count 10 nm to 1 μm	0.8	1	1 particle/cm^3	10 nm, <0.01 particles/cm^3
TSI FMPS, model 3091	UFP size distribution and count 5.6–560 nm	10	1	1 size distribution/sec	5.6 nm
TSI DustTrak, model 8520	PM2.5 mass	1.7	5	±0.001 mg/m^3	0.001–100 mg/m^3, 0.1–2.5 μm size range
Magee Scientific aethalometer	Black carbon	2.5–4	60	Proportional to flow rate	1 μg/m^3
Teledyne API, model 300E	CO	0.8	10	0.5% of reading	0.04 ppm
LI-COR, model LI-820	CO2	1	5	<2.5% of reading	3.0 ppm
RAE Systems, model PGM-7240	VOCs	0.4	<5	1 ppb	0–9999 ppb
Teledyne-API, model 200E	NOx, NO, NO2	0.5	20	0.5% of reading	0.4 ppb
Stalker LIDAR and vision digital system	Traffic documentation, distance and relative speed	NA	1	NA	NA

It should be mentioned that the instrument used to measure PM_2.5, the DustTrak, is not an ideal instrument for measuring diesel PM mass due to its lack of sensitivity to small particles found in diesel exhaust and distorted response to nonmonodispersed particles,27 though it does provide high time resolution signal. However, Anyon et al.28 showed a high correlation (R ² = 0.92) for the DustTrak with the gravimetric filter method (GFM) after developing needed sampling handling and preconditioning techniques. In this study, the DustTrak was not calibrated specifically for the measurement of particles emitted from diesel exhaust. Thus, PM_2.5 emission factors based on the DustTrak data might lack some accuracy for real-world emissions of HDDTs, although the precision of the data collected using the DustTrak still allows us to define the distribution of PM emission factors.

The instruments draw sample air from a 1.5-m sampling duct installed between the two passenger windows of the MMP. The sampling duct is equipped with fans to draw outside air in through an intake port in the right passenger window and out through the left passenger window. The residence time for air in the sampling duct is 0.3 sec. The instrument power supply, sampling manifold, and calibration procedures have been described previously.23,26

Sampling was conducted in Wilmington, CA, near the ports of Los Angeles and Long Beach. The MMP was driven on two routes during the study: the port/freeway/truck (PFT) route and the residential route. The PFT route covers primarily busy arterial roadways and was used to capture HDDT emissions, whereas the residential route covers more residential streets and was used to investigate emissions of LDGVs (see Figure 1). Both routes were about 30 miles long and driven two times per day (once in the morning between 8:00 and 10:30 and once in the afternoon between 14:30 and 17:00), two to three times per week, in the winter and summer seasons. Sampling was conducted in 2007 on February 10, 13, 20, 21, 23, 26, and 28 and March 1 and 4, and again on July 10, 13, 14, 17, 19, 25, 27, 29, and 31 and August 2, 6, and 9. A video camera recorded vehicles in front of the MMP and spoken observations were recorded covering off-screen events during each data collection event.

Figure 1. Sampling routes driven in winter and summer of 2007 (Port/Freeway/Truck route in yellow, Residential route in pink).

Emission Factor Measurements and Calculations

Measurements of vehicle exhaust plumes while driving behind or next to individual vehicles allowed for the determination of in-use fuel-based emission factors. The vehicle speed (above or below 30 mph) and engine operating condition (accelerating, idling, cruising) were assessed using the recorded MMP speed and video and audio records. Each calculated emission factor was categorized by driving condition defined as follows: The idle mode is defined as zero speed. The acceleration mode is defined as vehicle speeds greater than zero and acceleration greater than approximately 2 mph/sec (fast acceleration) or 1 mph/sec (slow acceleration) for longer than 3 sec. Cruising is defined as approximately steady-speed driving at speed greater or less than 30 mph.

Fuel-based emission factors were calculated based on concurrent increases in CO, CO₂, NO_x, BC, PM_2.5, or UFPs above the roadway concentrations, measured in the absence of the particular vehicle, normalized by carbon compound concentrations (CO₂, CO, and volatile organic compounds [VOCs]). In calculating the fuel-based emission factors, selecting the CO₂ baseline concentration is the major source of uncertainty. In this study, the baseline of each pollutant was defined as the value right before the MMP encountered a target vehicle and in the absence of other vehicles. Emission factors were calculated only for the vehicles that met with the criteria of no confounding sources nearby prior to and during at least 20 sec of chase period and a 40 ppm increase in CO₂ mixing ratio above the background. The criterion for CO₂ rise was set to ensure an appropriate plume capture and sufficient signal to calculate the emission factors with acceptable levels of uncertainty. That these criteria were met were verified by looking at the CO₂ concentrations prior to the plume capture and by video records to check for the proximity of the target vehicle and to verify the absence of other interfering plumes from other vehicles.

Achieving adequate plume capture from HDDTs proved very challenging, most likely due to the height of the exhaust stack relative to the MMP sample inlet at 1.5 m. In total, 260 HDDTs were chased, but only in 36%, or 93, of those cases were the criteria of no other vehicles in the vicinity and an increase of 40 ppm in CO₂ mixing ratio met to allow an emission factor calculation.

Fuel-based emission factors were expressed in terms of pollutant mass or particle number emitted per kilogram of fuel burned by normalizing the integrated concentration change in pollutant concentrations to total carbon concentration change, as shown in eq 1.9,15

(1)

where EF_P is the emission factor (grams of pollutant P per kg fuel burned), Δ[P] is the integrated change in mass concentration of pollutant P (μg m⁻³) above roadway concentrations for the duration of the plume capture; similarly, Δ[CO₂], Δ[CO], and Δ[VOC] are the integrated changes in the respective CO₂, CO, and VOC concentrations above the roadway concentrations with the VOC concentration (mg C m⁻³) expressed on a propane-equivalent basis. W _C is the carbon weight fraction of the fuel (0.85 for LDGVs, 0.87 for HDDTs).

Equation 2 was used to calculate the numbers of particles emitted per unit of fuel burned (particles/kg).9,15

(2)

where Δ[PN] is the integrated change in particle number concentration (particles/cm³) during the plume capture event.

An example of an exhaust plume capture from a LDGV is shown in Figure 2. As an individual exhaust plume was sampled, temporary increases in CO₂, CO, NO_x, VOCs, PM_2.5, UFP, and BC concentrations were observed. Note that each data point in Figure 2 was acquired by the data logger, which was set to record the concentrations from all instruments every 5 sec. The duration of the elevated level appears longer for NO_x due to a 20-sec response time, and longer for BC due to the 1-min response time, of the respective instrumentation measuring these species. We were typically not able to integrate the entire plume pulse duration for these pollutants because many plume captures were interrupted by other vehicles prior to pollutant concentrations returning to baseline values, especially for black carbon. Therefore, we calculated emission factors by integrating 20 sec of concentration excursion centered around the peak concentration measured for each pollutant. We performed a sensitivity analysis on integration times for emission factor calculations based on 5-, 10-, 15-, and 20-sec integration intervals. For pollutants measured by fast-response-time instruments such as CO, UFPs, and PM_2.5, their emission factors were relatively independent of integration time, showing less than 7% variation from their means. However, emission factors for NO_x and BC increased with longer integration times. Using a 5-sec integration time resulted in underestimating emission factors for NO_x and BC by 35% compared to using a 20-sec integration time. On the basis of the sensitivity analysis and peak widths, a 20-sec integration time was used to calculate the reported emission factors. Because the aethalometer had a 1-min response time, our calculated BC emission factors are expected to underestimate the actual BC values. Nevertheless, this approach still allows for the determination of relative BC emission factors and the identification of high emitters, as we will discuss in the next section.

Figure 2. Example of data obtained with MMP while following an individual LDGV (each point represents a 5-sec average value).

RESULTS AND DISCUSSION

Emissions Due to Different Engine Operations

Table 2 and Figure 3 summarize the results of all the emission factors calculated in this study categorized by vehicle type, i.e., LDGVs and HDDTs, and driving conditions (i.e., idling; accelerating at low speed and high speed; and cruising at low speed and high speed). In Figure 3, the black, blue, and red horizontal bars indicate median, mean, and 5 times the fleet-average values, respectively, for a particular vehicle type and engine operation combination. The box top and bottom show the 75th and 25th percentiles of emission factors in each category, respectively, and the outliers over the 95th or below the 5th percentile are shown as dots. Discrepancy between the median and means show that the emission factor data were skewed due to high-emitting vehicles. The values of red horizontal bars, 5 times the mean, were used as the criteria in this study to define a high-emitting vehicle. Extremely high emission factors in some LDGVs were observed for CO, NO_x, and BC during slow acceleration. These emission factors were up to 25 times higher than the average values. Among the 143 LDGVs, 6 vehicles were high emitters for CO and PM_2.5, 5 vehicles for NO_x, and 4 vehicles for UFPs. For the 93 HDDTs, one extremely high-emitting truck was identified during cruising at low speed. Its emission factors for NO_x, BC, PM_2.5, and UFPs were 17, 19, 20, and 12 times higher than the fleet-average values, respectively. Two HDDTs were high emitters for CO and BC, and five for NO_x, PM_2.5, and UFPs. Thus, approximately 5% of the fleet we measured consisted of high-emitting vehicles for both LDGVs and HDDTs.

Figure 3. Emissions due to different driving conditions. The black, blue, and red horizontal bars indicate median, mean, and 5 times the fleet-average values, respectively, for a particular vehicle type and engine operation combination. The box top and bottom show the 75th and 25th percentiles of emission factors in each category, respectively, and the outliers over the 95th or below the 5th percentile are shown as dots.

Table 2. Emission factors due to different driving conditions

		Idling		Slow Acceleration		Fast Acceleration		Cruise (<30 mph)		Cruise (>30 mph)
		LDGVs	HDDTs	LDGVs	HDDTs	LDGVs	HDDTs	LDGVs	HDDTs	LDGVs	HDDTs
CO (g/kg)	Median	8.1	10.7	11.9	8.1	6.0	3.2	13.7	5.3	7.2	15.5
	Mean	36.0	75.1	114.5	30.8	28.1	8.5	42.6	28.8	54.3	27.3
	5 × Mean	180.0	375.4	572.3	153.8	140.5	42.6	213.0	143.8	271.5	136.5
NOx (g/kg)	Median	4.0	15.5	3.9	6.8	3.5	14.4	5.7	7.5	2.5	17.6
	Mean	7.9	43.2	16.9	19.8	6.5	14.5	9.1	40.3	9.6	30.1
	5 × Mean	39.3	215.8	84.3	98.8	32.5	72.4	45.5	201.7	48.1	150.5
BC (g/kg)	Median	0.01	0.11	0.02	0.13	0.02	0.17	0.01	0.07	0.03	0.16
	Mean	0.04	0.22	0.09	0.17	0.06	0.17	0.05	0.97	0.09	0.22
	5 × Mean	0.21	1.08	0.46	0.87	0.32	0.87	0.27	4.87	0.45	1.12
PM2.5 (g/kg)	Median	0.04	0.19	0.07	0.33	0.06	0.10	0.05	0.15	0.05	0.08
	Mean	0.20	0.37	0.11	0.54	0.15	0.10	0.15	1.46	0.12	0.36
	5 × Mean	1.00	1.87	0.54	2.72	0.77	0.51	0.75	7.31	0.61	1.80
UFP (particles/kg)	Median	1.5 × 10^14	1.4 × 10^15	2.4 × 10^14	1.2 × 10^15	5.2 × 10^14	7.1 × 10^14	2.6 × 10^14	8.5 × 10^14	4.2 × 10^14	7.2 × 10^14
	Mean	9.6 × 10^14	3.3 × 10^15	5.0 × 10^14	4.2 × 10^15	6.6 × 10^14	6.9 × 10^14	5.2 × 10^14	3.6 × 10^15	5.5 × 10^14	1.4 × 10^15
	5 × Mean	4.8 × 10^15	1.6 × 10^16	2.5 × 10^15	2.1 × 10^16	3.3 × 10^15	3.5 × 10^15	2.6 × 10^15	1.8 × 10^16	2.7 × 10^15	7.1 × 10^15

In terms of fleet-average emission factors, HDDTs emitted much higher BC, PM_2.5, UFPs, and NO_x than LDGVs in each category as expected. Shah et al.13 reported fleet-average emission rates of CO, NO_x, and PM_2.5 for 11 on-road HDDTs using a mobile emission laboratory. They showed that per-mile emission rates for CO, NO_x, and PM_2.5 during creep, which simulates vehicle operation in heavily congested conditions, were higher than transient and cruise modes. Compared to the travel-based emission factors, the fuel-based emission factors showed less dependence on driving mode. This might be due to the lower sensitivity of fuel-based emission factors on different driving modes, as reported by Kean et al.29. A previous study showed that particle formation of nucleation mode occurs from diesel engines without diesel particle filters at low engine load.30 This may explain why our results showed that on a per-fuel basis, the highest PM_2.5 and UFP emission factors occurred during slow acceleration or cruising at low speed. For LDGVs, we observed CO, NO_x, and BC emissions to be slightly higher during idle or slow acceleration, similar to what has been reported in other studies.31

Concerns have been expressed that excess emissions during cold-start operation may contribute a considerable fraction of total hydrocarbon, CO, and NO_x emissions from the motor vehicle fleet.21,32,33 Singer et al.32 compared fuel-based hot-stabilized emissions and cold-start emissions from LDGVs under in-use conditions. They showed that cold-start emissions of nonmethane hydrocarbons, CO, and NO_x were about 3 times higher than stabilized exhaust emissions. Cold-start emissions of elemental carbon and particle number from LDGVs were 1.9 and 3.3 times higher than hot-start emissions in a study by Kittelson et al.21 Lough et al.34 also investigated the impact of cold start on vehicle emission rates of nonmethane hydrocarbons (NMHCs) by collecting samples as vehicles exit a parking lot in subzero temperatures. They found that NMHC emissions during cold start were about double the emissions measured in a tunnel, where the vehicles would have driven long enough to have reached optimal operating temperatures. Because the high-emitting vehicles identified here emitted 5–25 times more than the mean values, they were relatively easy to identify, and confounded by neither the variability introduced due to cold starts nor driving cycle effects.

Emission Distributions

Frequency distributions of individual vehicle CO, NO_x and PM_2.5 emission factors for LDGVs and HDDTs are shown in Figures 4 and 5. The emission factors for each pollutant showed significant variation, likely due to the complex dependence of vehicle emissions on factors such as vehicle age, engine type, maintenance, and driving conditions. In addition, as the emission standards become very stringent, significant variation in vehicle emissions is expected. The Low Emission Vehicle (LEV) II programs, for example, require reductions of 98% for hydrocarbon (HC), 96% for CO, and 98% for NO_x when compared to emissions from older passenger cars that were not subject to emission standards. For the verification of our results, the arithmetic means of all pollutant emission factors were calculated and compared to emission values from other studies, as shown in Table 3.

Table 3. Comparison of emission data from current study with previous studies

Reference	Vehicle Type	Location, Year	Approach	CO (g/kg)	NOx (g/kg)	BC (g/kg)	PM2.5 (g/kg)	UFP (particles/kg)
This work	LD	Wilmington, CA 2007	MMP	47	9.4	0.06	0.15	6.0 × 10^14
	HD	Wilmington, CA 2007	MMP	36	34	0.5	0.73	4.5 × 10^15
Kirchestetter et al. (1999)^15	LD	Caldecott Tunnel 1997	Tunnel		9.0	0.04	0.11	4.6 × 10^14
	HD	Caldecott Tunnel 1997	Tunnel		42	1.3	2.50	6.3 × 10^15
Geller et al. (2005)^16	LD	Caldecott Tunnel 2004	Tunnel			0.02	0.07	2.5 × 10^14
	HD	Caldecott Tunnel 2004	Tunnel		40	0.9	1.02	8.2 × 10^15
Ban-Weiss et al. (2008)^9	LD	Caldecott Tunnel 2006	Tunnel		3.0	0.02	0.07
	HD	Caldecott Tunnel 2006	Tunnel		40	0.9	1.40
Kuhns et al. (2004)^35	LD	Las Vegas, NV 2000–2002	Remote sensing	49	8.8 ^a		0.06
	HD	Las Vegas, NV 2000–2002	Remote sensing	10	19.9 ^a		1.50
Bishop and Stedman (2008)^10	LD	West LA, CA 2005	Remote sensing	28	3.2 ^a
Thornhill et al. (2010)^37	LD	Mexico City	Mobile lab with PMF ^b	251	3.3	0	0.04
	HD	Mexico City	Mobile lab with PMF ^b	28	38	1.4	0.37
Notes: ^aEmission factors reported as NO; ^bPositive matrix factorization receptor modeling.

Figure 4. Histograms of CO, NO_x, and PM_2.5 emission factors from 143 individual LDGVs in Wilmington, CA. Only 133 vehicles were included for CO. Average emission factors for the highest 5% of emitting vehicles, and EMFAC emission factors in g/mi were converted to g/kg using a fuel consumption rate of 10 L/100 km for LDGVs.

Figure 5. Histograms of CO, NO_x, and PM_2.5 emission factors from 93 individual HDDTs in Wilmington, CA. Only 91 vehicles were considered for CO. Average emission factors for the highest 5% of emitting vehicles, and EMFAC emission factors in g/mi were converted to g/kg using a fuel consumption rate of 49.5 L/100 km for HDDTs.

The calculated LDGV fleet-average emission factors agreed well with previous remote sensing10,35 and tunnel studies in the United States.9,15,16 However, the emission characteristics from the vehicle fleet are significantly different between Mexico City and U.S. cities.36,37 Comparison of CO emissions showed that CO fleet-average emission factor of LDGVs in Mexico City was over 3 times higher than ones in U.S. cities (Wilmington, Las Vegas, and Los Angeles) due to the older vehicle fleet in Mexico City, as shown in Table 3.

In addition, EMFAC, a statistical model of on-road vehicle emissions,38 was also used to compare our results with predictions of the fleet-average emissions for Los Angeles county. Our emission factors for CO and NO_x were higher than EMFAC predictions for Los Angeles County, which could be indicative of an older vehicle population in Wilmington, but emission factors of PM_2.5 were in good agreement.

As mentioned earlier, we expect our BC emission factors to be underestimated due to the 1-min aethalometer response time. Interestingly, the fleet-average BC emission factor for LDGVs was slightly higher than the results from other research.9,15 This may be caused by a few very-high-emitting vehicles. Their emissions dominated the fleet-average result and may be indicative of a higher occurrence of BC high emitters in Wilmington, a low-socioeconomic-status community that would be expected to have a greater than average proportion of older, poorly maintained vehicles.39,40 For instance, our overall BC emission factor for LDGVs was 0.06 g/kg, yet the value for the highest 5% was 0.4 g/kg and the fleet-average emission factor excluding the highest 5% vehicles was 0.04 g/kg, which agreed well with other research.15 The average emission factors for the top 5% high-emitting vehicles were 7–18 times higher than the total vehicle-average emission factors for all pollutants.

For HDDTs, fleet-average emission factors were also in reasonable agreement with previous tunnel, remote sensing, and mobile laboratory chase studies.9,10,15,16,35,37 According to Ban-Weiss et al.41 who measured emission factors for BC and particle number (PN) from 226 individual HDDTs in Caldecott Tunnel, CA, the mean emission factors of BC and PN were 1.7 g/kg and 4.7 × 10¹⁵ particles/kg, respectively. Our fleet-average UFP emission factor of HDDTs agreed very well with their results, whereas our BC emission was lower due to the slow response time of the aethalometer.

For the EMFAC model comparison of HDDTs, CO emission factors were higher than model predictions from Los Angeles County, whereas emission factors for NO_x and PM_2.5 were lower. Averaged emission factors for CO and NO_x for the top 5% high-emitting HDDTs were about 11 times higher than total vehicle-averaged factors. The emission factor for the top 5% of BC high-emitting HDDTs was 18 times higher than the overall fleet average. As expected, the BC fleet-average emission factor for HDDTs was lower than values obtained in previous tunnel studies.9,15,41

Contributions of High Emitters to Total Fleet-Average Emission Factors

Contributions of individual vehicles to the overall vehicle-average emission factor for individual pollutants are shown in Figure 6. The x-axis represents the number of vehicles excluded when calculating fleet-average emission factors. Vehicles were excluded from the calculation starting with the most polluting vehicle and ending with the least polluting vehicle. The y-axis shows the ratio of the fleet-average emission factor after excluding the corresponding vehicles to the total vehicle-average emission factor. The skewness of the resultant graph indicates the degree to which the calculated fleet-average emission factor is affected by high-emitting vehicles in the sampled fleet. The difference between the mean of all vehicles and the mean of vehicles after excluding each vehicle illustrates the large impact a small fraction of the vehicle fleet has on the fleet-average emission factor for the current data set. For LDGVs, if the highest 5% of emitting vehicles are removed, the average emission factors for NO_x, BC, PM_2.5, and UFPs are reduced by 34%, 39%, 44%, and 31%, respectively, whereas the CO emission factor is reduced by 50%.

Figure 6. Emission contribution of individual vehicles to the overall fleet-average emission factors.

Emission distributions of HDDTs are significantly more skewed than for LDGVs. Ban-Weiss et al.41 showed that the highest-emitting 10% of trucks contributed about 40% of total BC and PN emissions from all HD trucks. Our results from 93 HDDTs in Wilmington were more skewed compared to this Caldecott Tunnel study. The fleet-average emission factors of PM_2.5, UFPs, and BC would be reduced by 60% and about 50% for NO_x and CO without the highest-emitting 5% HDDTs. Note that CO is the most skewed pollutant for LDGVs, whereas BC is the most skewed for HDDTs. Most observed high-emitting vehicles were older and their exhaust plumes were visible. It is important to note that the extent to which these high-emitting vehicles are responsible for a disproportionate amount of the overall emissions is dependent on how much they are used compared to the rest of the fleet.

In Figure 7, emission factors for the highest 10% of BC emitters were plotted against matched NO_x and PM_2.5 to investigate relationships between these pollutants for high-emitting vehicles. Although our results showed that about 5% of the fleet were high emitters, the top 10% of high-emitting vehicles were chosen to examine overlaps between high BC, NO_x, and PM_2.5 emissions and provide sufficient data for a statistically significant comparison. Among the top 10% of BC emitters in LDGVs, 43% and 29% were also in the highest 10% of PM_2.5 and NO_x emitters, respectively. In HDDTs, there was significant overlap between the top 10% high BC, PM_2.5, and NO_x emitters. Of the top 10% of BC-emitting HDDTs, 56% and 78% were also in the top 10% NO_x and PM_2.5 emitters, respectively. This suggests that a repair/retrofit/replacement program aimed at the highest-emitting HDDTs could concurrently reduce BC, NO_x, and PM_2.5 emissions.

Figure 7. Emission factors for the highest 10% of BC emitters plotted against matched NO_x and PM_2.5. Boxes highlight the highest 10% of emitters for each pollutant: solid box for PM_2.5, dotted box for NO_x.

CONCLUSIONS

The present study has demonstrated the advantages of using a zero-emission MMP for on-road emission measurements, including the ability to sample individual vehicles under different in-use operating conditions, and to characterize high-emitting vehicles emissions based on a large vehicle sample. Approximately 5% of the 143 LDGVs and 93 HDDTs measured in the Wilmington area were high emitters, whose emissions were more than 5 times the fleet-average values. These high emitters have a strong effect on the overall fleet emissions. If high emitters are excluded, the average emission factor is reduced by 30% and more than 50% for LDGVs and HDDVs, respectively. Nevertheless, the current study was still limited to vehicles from a single location in determining the extent to which high emitters contribute to overall LDGV and HDDT emissions. Investigating a larger data set, including vehicles from other geographical and socioeconomic locations, will provide a more complete characterization of on-road vehicle emissions. It is possible that as the overall fleet becomes cleaner due to more stringent regulations, a small fraction of the fleet may contribute a growing and disproportionate share of the overall on-road vehicle emissions. Therefore, long-term changes in on-road emissions need to be monitored, especially for the characterization of high-emitting vehicle emissions.