An Experimental Framework of Particulate Matter Emission Factor Development for Traffic Modeling

Abstract

To estimate traffic facility-oriented particulate matter (PM) emissions, emission factors are both necessary and critical for traffic planners and the community of traffic professionals. This study used locally calibrated laser-scattering sensors to collect PM emission concentrations in a tunnel. Emission factors of both light-duty and heavy-duty vehicles were found to be higher in autumn compared to summer. Based on this study’s data analysis, PM emissions, in terms of mass, have a strong seasonal effect. The study also conducted a PM composition test on normal days and during haze events. Preliminary results suggested that the transformation of gaseous tailpipe emissions to PM is significant within the tunnel during a haze event. This study, therefore, recommends locally calibrated portable devices to monitor mobile-source traffic emissions. The study suggests that emission factor estimation of traffic modeling packages should consider the dynamic PM formation mechanism. The study also presents traffic policy implications regarding PM emission control.

1. Introduction

Particulate matter (PM) pollution is a hot-button issue in environmental and traffic studies. A considerably complicated compound of fine particles, PM also comprises liquid droplets such as soot, ash, hydrocarbons (HC), and water-soluble organic fractions [1]. Vehicle emissions are a major source of PM pollution, including emissions from vehicle tailpipe exhausts and brake wear, as well as road dust. Among these sources, fuel combustion and emissions through vehicle engine exhausts emit a large number of fine particles (PM_2.5) [2]. Techniques for measuring airborne particles have been developed and have evolved over a long period of time. The two main methods for PM_2.5 monitoring are the manual sampling and weighing method and the continuous automatic method. The latter includes the tapered element oscillating microbalance (TEOM) and the beta-attenuation monitor (BAM). Moreover, PM size distribution and topography can be measured by a scanning mobility particle sizer (SMPS) and scanning electron microscopes (SEMs), respectively [3]. Analysis of PM chemical composition is also sophisticated through the use of devices such as inductively coupled plasma optical emission spectroscopy (ICP-OES) for metal elements, the thermal optical carbon analyzer for carbonaceous aerosols, gas chromatography–mass spectrometry (GC-MS) for organics, and ion chromatography for water-soluble ions [4]. Some of these devices have been deployed to investigate the traffic emission component of air pollution [5,6]. Thus, measures, such as the usage of vegetation and noise barriers, have been implemented to mitigate the air pollution exposure level of residents who live close to traffic routes [7]. Traffic engineering tends to focus on the link between traffic activities and PM emissions. At the same time, macroscopic traffic software delivers aggregated information at the road network level, such as road link volume, traffic composition, and average speed. Estimations of PM emission factors, on the basis of vehicle kilometers traveled (VKT) and fuel consumption, are therefore critical for the macroscopic software used by traffic planners and the community of traffic professionals. However, PM_2.5 is not entirely emitted by vehicles on traffic routes or generated by precursors emitted by vehicles. This study adopts the integrated engineering approach to employ some environmental technologies in traffic engineering [8]. This PM study specifies its research object as the main PM emissions by traffic motorway facilities. These comprise the fine emissions of tailpipe PM and PM precursors, followed by the varied mechanisms of PM_2.5 formation processes at road traffic facilities. The study then delivers an estimation of the PM_2.5 emission factor with its magnitude of uncertainty. Moreover, it proposes a new framework for roadway facility PM_2.5 estimations considering seasonal factors, while also discussing the impact of environmental events on PM formation at roadway facilities. Finally, the study discusses the PM control implications for traffic policy.

Traffic-induced emissions are one of the sources of PM_2.5 and a major PM_2.5 emission contributor in urban areas. The three categories of traffic PM emission studies defined by traffic system elements are, firstly, vehicle-oriented studies. These include chassis and onboard tests, with studies also conducted to associate disaggregated vehicular operations with instantaneous tailpipe emissions [9,10]. Secondly, human-oriented pedestrian or passenger exposure studies are extensively implemented to estimate the adversary impacts on temporal and spatial scales [11]. Besides large-scale monitoring sensor networks, some studies employ artificial intelligence (AI) to estimate ambient concentrations and predict haze hazards [12,13]. The entire volume of PM_2.5 at traffic facilities is not necessarily emitted or even induced by vehicles. Thirdly, traffic facility-oriented studies on tunnels and roadsides are also implemented to estimate emission factors [14]. In summary, PM induced by internal combustion engine (ICE) vehicles can be categorized into three sources:

• Particulate matter (PM) is produced by the combustion of fossil fuels during the driving of ICE vehicles. The formation process of this proportion of PM is relatively complicated but is generally considered to proceed with three steps: the cracking of fossil fuels, nucleation in the air, and PM agglomeration and oxidation. In this process, fossil fuels generate a large amount of organic matter and gaseous precursors that grow into larger particles through agglomeration or chemical changes to generate new particles [15].
• Particulate matter (PM) is generated by friction between parts (e.g., metal gears, rubber parts, etc.) during the driving of ICE vehicles. These particles mainly exist in the form of metal or organic particles and enter the atmosphere [16,17].
• Resuspended road dust generated during the driving of ICE vehicles is also one of the main sources of atmospheric PM, but its composition is relatively simple and mainly consists of crustal elements (e.g., silicon [Si] and magnesium [Mg]) [18].

Particulate matter (PM) can be categorized into primary PM (PPM) and secondary PM (SPM). The former consists of chloride, hydrocarbon-like organic aerosols (HOA), cooking OA (COA), and coal combustion OA (CCOA). The latter includes sulfates (SO₄^2–), nitrates (NO₃^–), ammonium sulfate (NH₄⁺), and oxygenated OA (OOA). Secondary particulate matter (SPM) is a significant contributor to the accumulation of PM concentration during haze events [19]. The current traffic facility-oriented PM study specifically studies the primary tailpipe PM_2.5 and secondary PM_2.5 induced by tailpipe gaseous precursors at roadsides. Unlike the emission–dispersion procedure of gaseous emissions, traffic-induced PM largely follows the emission–formation–floating (or deposition, depending on PM size) associated with dynamic physical–chemical reactions. In contrast to the emission–dispersion development framework of the gaseous emission factor, the proposed new framework takes the formation procedure into consideration.

In addition to the three direct PM emissions identified above, ICE vehicles also produce a large number of gaseous substances, such as nitrogen oxides, sulfides, hydrocarbons, and ammonia. These substances, as PM precursors, undergo chemical reactions to generate large amounts of secondary PM which significantly affect the PM mass concentration in the road area.

The tailpipe PM emissions of mobile vehicles are spatially distributed, with the PM mass concentration level highly dynamic on a temporal scale. In roadside PM monitoring, vehicle emissions contribute to creating unique seasonal variation characteristics in PM concentration in this area. This is also significantly different from other emission sources in its temporal and spatial fluctuations. Beta-attenuation monitors are the standard equipment used for mass PM concentration measurement in environmental studies [20]. However, these PM sensing devices are not cost-effective, as the low-affordability station is bulky and has low mobility to deal with vehicle emission sources [21]. Hence, in a facility-oriented PM study, PM measurement equipment should respond to the study’s requirements. The portable Koala PM sensor is considered satisfactory as the laser-scattering device is financially affordable and reliable after it is calibrated locally [22]. The goodness of fit with the beta-attenuation equipment can reach approximately 0.9 after calibration. This sensor can collect data on temperature, humidity, mass concentration, and size distribution [22].

In contrast to the findings of studies on gaseous traffic pollutants, the formation of PM emissions is significantly affected by regional transportation and meteorological conditions. Research on air pollution events during the city-wide traffic lockdown for the COVID-19 pandemic in Hubei Province, China, demonstrates that half the pollution episodes were associated with the regional transportation of air pollutants from upstream sources. Local pollution was a major cause of the remaining episodes due to stagnant meteorological conditions and low planetary boundary layer (PBL) height [19]. Therefore, the PM emissions collected at the roadside were not entirely generated in the vicinity.

Vehicle tailpipe emissions are emitted and follow varied formation mechanisms through physical–chemical processes. Compared to gaseous emissions, PM generated by ICE vehicles has volatile characteristics, not only in particle size distribution but also in chemical composition. Various road traffic factors substantially affect the concentration, composition, and particle size distribution of PM emitted by ICE vehicles. The main factors are the composition of road vehicles (e.g., the ratio of gasoline engine vehicles to diesel vehicles, the number of vehicles with various emission standards, etc.), the operating conditions of ICE vehicles (e.g., vehicle age, driving style, etc.), and overall road conditions (e.g., road slope, roughness, etc.). At the same time, environmental factors, such as atmospheric acidity and oxidization, ambient temperature, ambient humidity, light intensity, etc., also affect the contribution of particulate matter (PM). Fine particles go through nucleation, agglomeration, reaction, and other processes and finally enter the atmosphere [23].

Some gaseous substances or other particles in the atmosphere are chemically converted into new particles. These newly generated particles are secondary particles, with these gaseous substances or particles known as precursors of secondary particles. Traffic emissions produce gaseous precursors, such as sulfur dioxide (SO₂), nitrogen oxides (NO_x), and ammonia (NH₃). Sulfur dioxide (SO₂) is not only one of the main precursors of SPM but also one of the primary air pollutants in China. Its main source is fossil fuel combustion and processing of other sulfur-containing substances. In transportation, diesel engines may run on high-level sulfur fuel, increasing SO₂ emissions. Nitrogen oxides (NO_x) are air pollutants and also precursors of secondary particulate matter (SPM). One of their main sources is road vehicles. As a precursor of SPM, NO_x generates photochemical smog through photochemical oxidation under certain environmental conditions to generate secondary particulate matter (SPM) [24]. Due to the continuous increase in urban car ownership, the anthropogenic emission sources of NO_x produced by ICE vehicles in megacities such as Beijing and Shanghai have significantly increased. Ammonia (NH₃) is also an air pollutant and a precursor of secondary particulate matter (SPM). In traffic sources, NH₃ mainly comes from the reaction between three-way catalytic devices and NO_x from exhaust gas emissions and fossil fuel combustion [25]. As the source of NH₃ emissions is mainly ground emissions, they usually stay on the ground for a short period.

After PM is emitted via vehicle tailpipes, it undergoes a formation procedure. With increasing PM diameter, PM deposits rather than floats in the atmosphere. Hence, the scope of the current study comprises only the volatile tailpipe primary PM (PPM) and the secondary PM (SPM) induced by vehicle gaseous precursors. The occurrence of haze events is highly associated with meteorological conditions. According to the Chinese Meteorological Administration (CMA) definition, a haze event must meet the following requirements, that is, visibility < 10 km and relative humidity (RH) < 90%. Considering the CMA definition and this study’s measured results, a haze event is further clarified in this study when the average concentration of PM_2.5 is above 75 micrograms (μg)/m³ and RH < 90%. One study of air pollution events focused on traffic emissions as Wuhan was in regional lockdown with barely any road traffic due to the COVID-19 pandemic in Hubei Province, Central China. The study found that 50% of pollution episodes were associated with the long-range transportation of air pollutants from upstream source regions, while local pollution was dominant for the remaining episodes owing to stagnant meteorological conditions [26]. Meteorological conditions regulate the PM haze events in Beijing. While the chemical composition of PM in Beijing is similar to other findings worldwide, gaseous emissions of volatile organic compounds (VOCs), NO_x, and SO₂ are responsible for large volumes of SPM formation [27].

Tunnel studies have been implemented extensively across many countries [28]. Road tunnels can largely mitigate photolysis actions with their insufficient amount of sunlight, thus collecting primary vehicular emissions. Although road tunnel studies can provide limited traffic operational status, they can provide a better opportunity to collect particles due to the limited meteorological effects occurring inside tunnels [29,30]. In summary, this study is focused on PPM emitted by vehicles and tailpipe gas precursors that lead to SPM to estimate dynamic PM_2.5 emissions of road facilities under varied meteorological conditions and to discuss the traffic policy implications of PM_2.5 mitigation. Therefore, in this tunnel study, undertaken to minimize ambient effects, portable laser scattering devices are used for the measurement of PM mass. The main objective of this study is thus to estimate the real-world average emission factors (EFs) of PM_2.5 from on-road vehicles by vehicle category in the metropolitan area of Beijing, China.

2. Methodology

2.1. Regression Analysis

Tunnel tests are common practice for PM studies. Due to the significant variations in PM size distribution and composition at different locations, numerous studies have been conducted to estimate the PM emission factors.

The basic principle of the tunnel test method is the principle of mass conservation. The experimental tunnel is regarded as a closed columnar space, and the total mass of PM generated by ICM vehicles in the tunnel can be obtained from the mass difference of PM measured by the PM sensors installed at the tunnel’s ingress and egress. The model must be based on the following premise: the tunnel has only two vents at its ingress and egress with no other ventilation facilities at the chosen site [23].

The calculation formula is as follows:

$$E=\left(C_{out}\times V_{out}\right)-\left(C_{in}\times V_{in}\right)$$

(1)

$$V_{out}={\mathrm{S}}_{out}\times W_{out}\times t$$

(2)

$$V_{in}={\mathrm{S}}_{in}\times W_{in}\times t$$

(3)

$E$ = Total emissions of PM from all ICE vehicles in the tunnel (mg)

$C_{out}$ = Mass concentration of PM measured at the tunnel exit (egress) (mg/m³)

$C_{in}$ = Mass concentration of PM measured at the tunnel entrance (ingress) (mg/m³)

$V_{out}$ = Ventilation volume calculated at the tunnel exit (egress) (m³)

$V_{in}$ = Ventilation volume calculated at the tunnel entrance (ingress) (m³)

${\mathrm{S}}_{out}$ = Cross-section area of the tunnel exit (egress) (m²)

${\mathrm{S}}_{in}$ = Cross-section area of the tunnel entrance (ingress) (m²)

$W_{out}$ =Average wind speed at the tunnel exit (egress) (m/s)

$W_{in}$ = Average wind speed at the tunnel entrance (ingress) (m/s)

$t$ = sampling duration (s).

It is assumed that no significant environmental differences exist between the tunnel entrance and exit. The cross-section area of the tunnel is assumed to be constant. The formula is therefore simplified to calculate the total PM mass, with the simplified formula shown in Equation (4):

$$E=\left(C_{out}-C_{in}\right)\times S\times W\times t$$

(4)

The least-squares regression analysis has been extensively applied for source appointments and emission factor estimation with these estimated by Equation (5):

$$E_{\mathrm{average}}=\frac{E}{N\times L}$$

(5)

where

$E_{\mathrm{average}}$ = Average emission factors for all vehicles passing through the tunnel (mg/(veh·km)

$N$ = Total traffic in the tunnel (veh)

$L$ = Mileage traveled by the vehicle: the length of the tunnel (km).

By introducing the number of vehicles of different types, a multiple regression model is established, and the particle emission factors of different categories of vehicles are obtained through parameter estimation:

$$E_n=X_{n\times p}\times {\beta }_p+{\epsilon }_n$$

(6)

$${\beta }_p={\mathrm{EF}}_p\times L$$

(7)

where

$E_n$ = In the n^th experimental period, the total emissions of all motor vehicles (mg)

$X_{n\times p}$ = In the n^th experimental period, the number of vehicles of type p (veh)

${\beta }_p$ = Amount of PM generated by the p^th type of vehicle (mg/veh)

${\mathrm{EF}}_p$ = PM factors for the p^th type motor vehicles (mg/(veh·km)

$L$ = Length of the experimental tunnel (km)

${\epsilon }_n$ = Error term.

Therefore, the estimation of the coefficients is shown in Equation (8).

$${\mathsf{\beta }}_{\mathrm{p}}={\left({\mathrm{X}}^{\prime }\mathrm{X}\right)}^{-1}{\mathrm{X}}^{\prime }\mathrm{E}$$

(8)

The estimate of the residual variance for the model is as follows:

$${\sigma }_{error}^2=\frac{{\mathsf{\epsilon }}^{\prime }\mathsf{\epsilon }}{n-p}$$

(9)

The covariance matrix of the coefficients is as follows:

$$\mathrm{Var}\left(\mathsf{\beta }\right)=\frac{{\mathsf{\epsilon }}^{\prime }\mathsf{\epsilon }}{\mathrm{n}-\mathrm{p}}{({\mathrm{X}}^{\prime }\mathrm{X})}^{-1}$$

(10)

Thus, the standard errors of the ${\mathsf{\beta }}_{\mathrm{p}}$ coefficient can be found by computing the square roots of the diagonal elements in the variance–covariance matrix of the coefficients in Equation (10).

2.2. PM Sampling and Composition Analysis

In Beijing, in the recent decade, haze events were very active from late autumn to early spring. Peak PM_2.5 concentrations were higher than 160 μg/m³ during the experiment. At the same time, traffic volume and vehicle composition have not dramatically changed. The regression model yielded biased results that underestimated total emissions, suggesting that PM formation has significantly changed within the tunnel. Therefore, composition analysis of the in-situ samples was conducted to explore the possible formation mechanism(s) and the related role of vehicle-emitted gaseous pollutants.

3. Field Experiment

The tunnel, selected as the experimental site, is in the southern metropolitan area of Beijing. The tunnel is a 305 m, 3-lane straight single-way straight road with neither a steep uphill nor a steep downhill gradient. The sampling sites are located at the tunnel egress (exit), ingress (entrance), and background site. No substantial change is found in the cross-section of the entire tunnel, shown in Figure 1.

The experiment used Koala PM sensors, an anemometer, a video camera, and Doppler radar, with monitoring points set up at both the tunnel entrance and exit. The PM sensors were locally calibrated to accustom them to the PM composition in Beijing, demonstrating good fit and reliability as mentioned above.

The devices are illustrated in Figure 2. The in-situ frequency-modulated continuous wave (FMCW) Doppler radar collected data on vehicle speed, vehicle category, and traffic volume [31], while the video data checked traffic volume by the varied vehicle categories. Electric vehicles were distinguished from ICE vehicles by the car number plate color after the field experiment. The Koala sets collected the mass and quantity concentrations of six different particle sizes, with a temporal resolution is 5 min (min). To smooth out fluctuations, all types of data were aggregated at the hourly level. The experiment also had temperature and humidity sensor modules. The purpose of the experiment was to obtain the difference in PM mass concentration at the tunnel entrance and exit.

Three portable PM sensors were mounted at each monitoring point to reduce data error caused by instrumentation and to ensure monitoring quality. The homogeneity of the PM data was checked by the Kolmogorov–Smirnov test. The average concentrations of PM at the tunnel entrance and exit were used to ultimately calculate the total amount of PM generated by the vehicle when passing through the tunnel. The experiments were conducted in two seasons, from September–November 2020 and from May–June 2021, with the daily experimental period from approximately 9:00 to 16:00. In total, 174 h of data were collected, with valid data of 64 h in autumn and 39 h in summer. Any data involving incidents were screened out. In autumn, 15 h of haze events occurred while, in summer, 6 h of high humidity were recorded. The experiment’s equipment installation is shown in Figure 2.

Samples were collected by the system shown in Figure 3. These samples were as follows: PM_2.5 samples collected on polytetrafluoroethylene (PTFE, Teflon) filters (2 µm pore size; 46.2 mm diameter, Whatman North American Headquarters, Florham Park, NJ, United States) by a small volume (5 L/min) air sampler (LD-6C, Beijing Greenwood Environmental Technology Co., Ltd., Beijing, China). The LD-6C was a laser light-scattering dust monitor with a built-in pump for particle sampling. All devices were arranged on the sidewalk, with the sidewalk roughly 1.6 m above the pavement. Depending on each sampling condition, the sampling timespan was from 5–8 h. The Teflon filters were pre- and post-weighed with a 0.1 mg resolution analytical balance. Afterward, they were stored in a box at constant temperature and humidity for use. The blank filters in the field were opened and collected at the sampling sites.

Analysis Equipment

The PM_2.5 filter samples were ultrasonically extracted twice by ultrasonic diluting for 20 min, using deionized water, and filtered by disposable aqueous microporous membranes (pore size of 0.45 µm). Ten (10) major water-soluble inorganic ions (anions: Cl⁻, F⁻, NO₂⁻, NO₃⁻, and SO₄²⁻; cations: NH₄⁺, Li⁺, Na⁺, K⁺, and Mg²⁺) were measured by ion chromatography (ICS5000+, Thermo Scientific™ Dionex™, Waltham, MA, USA). The PM_2.5 sample solutions were exchanged and separated by the anion or cation chromatography column and detected by an inhibited conductivity detector. The qualitative analysis was conducted based on retention time, with the quantitative analysis conducted based on peak height or peak area. The field and laboratory blank filters were processed using the same methods and measured simultaneously. The mass concentrations of water-soluble ions were calculated according to the following formula:

$$\rho =\frac{\left({\rho }_1-{\rho }_0\right)\times V\times N\ast D}{V_{nd}}$$

(11)

where

$\rho$ = Mass concentration of the water-soluble ions in the PM_2.5 filter samples

${\rho }_1$ = Mass concentration of the water-soluble ions in the extracted dilution of the filter samples (mg/L)

${\rho }_0$ = Mass concentration of the water-soluble ions in the extracted dilution of the blank filter samples (mg/L)

V = Volume of the extracted dilution (mL)

N = Number of cut portions of the filter membrane, if the whole filter was used, then N = 1, if ¼ of the filter was used (N = 4)

D = Dilution ratio

$V_{nd}$ = Total sampling volume under standard state (101.325 kPa, 273K).

4. Data Analysis and Discussion

4.1. Estimation of Emission Factors

The detection of motor vehicles and the measurement of spot speed during the experiments were completed using Doppler radar. In Beijing, private diesel-powered light-duty vehicles (LDVs) are prohibited from entering metropolitan areas (within the 5th loop line), whereas diesel-powered heavy-duty vehicles (HDVs), for example, buses and trucks, can access those areas. Electric vehicles (EVs) were screened out by their distinct number plate color due to their low PM emission intensity [32]. Hence, Figure 4 shows the hourly traffic volume of diesel-powered heavy-duty vehicles (HDVs) and petrol-powered light-duty vehicles (LDVs) during the experimental period.

The average hourly flows of LDVs and HDVs were 702 and 17, respectively. The average hourly flows of LDVs and HDVs on weekdays were 698 vehicles/hour and 19 vehicles/hour, respectively, while the average hourly flow of LDVs and HDVs on weekends reached 706 vehicles/hour and 16 vehicles/hour, respectively. The maximum hourly flows on weekdays of LDVs and HDVs reached 939 vehicles/hour and 40 vehicles/hour, respectively, while the maximum hourly flows of LDVs and HDVs on weekends reached 1022 vehicles/hour and 36 vehicles/hour, respectively.

The volume of HDVs showed little change throughout the experimental period, as most of them, for instance, buses and recycling trucks, were operated on a routine basis. However, the number of private LDVs varied greatly between the peak period and lowest level, and traffic flows on weekends were greater than on weekdays.

Table 1. Traffic Statistics during the Experiment.

	Summer Average Volume (Veh/h)	Summer Max Volume (Veh/h)	Autumn Average Volume (Veh/h)	Autumn Max Volume (Veh/h)
LDV	704	939	701	1022
HDV	18	40	16	36

shows the speed data.

During the experiment, the time series of hourly average speed were analyzed. The average speed of all motor vehicles was about 58.8 km/h, with no significant difference between weekdays and weekends with average speeds of 59.5 km/h and 58.4 km/h, respectively. The average speeds in summer and autumn were 57.7 km/h and 60.1 km/h, respectively. Therefore, no significant difference was found in speeds between summer and autumn.

A multiple regression model was established, with the total amount of PM produced by motor vehicles as the dependent variable. The independent variables were (1) the two types of motor vehicles (i.e., LDVs and HDVs) and (2) the environmental temperature, as shown in Equation (12).

$$E=L\times E{F}_{LDV}\times Q_{LDV}+L\times E{F}_{HDV}\times Q_{HDV}+\beta \times T$$

(12)

where

${EF}_{LDV}$ = LDV PM emission factor (mg/[veh·km])

$Q_{LDV}$ = LDV traffic count (veh)

${EF}_{HDV}$ = HDV PM emission factor (mg/[veh·km])

$Q_{HDV}$ = HDV traffic count (veh)

$\beta$ = Coefficient

$T$ = Temperature.

Therefore:

$$E=0.89493\times Q_{LDV}+29.1018\times Q_{HDV}+(-7.133)\times T$$

(13)

Likewise, the PM factors of LDVs and HDVs in autumn were calculated in Equation (14):

$$E=2.258\times Q_{LDV}+45.095\times Q_{HDV}+(-61.396)\times T$$

(14)

To verify the established PM emission model in this study, the goodness of fit (R² values), t-tests, and F-tests were used. For the regression model in summer, the p-values of the LDV and HDV coefficients were substantially less than 0.05, while the p-value of the temperature coefficient was slightly higher than 0.05. For the regression model in summer, all coefficient estimations were statistically significant.

As shown in

Table 2. Emission Factors in Summer and Autumn.

Season	LDV Emission Factor mg/(veh.km)	HDV Emission Factor mg/(veh.km)
Summer	2.9 ± 0.7	95.4 ± 10.1
Autumn	7.0 ± 1.9	147.9 ± 80.2

, the PM emission factors of HDVs were found to be significantly higher than those of LDVs. The main reason for the emission factor gap between diesel vehicles and gasoline vehicles was the working principles of the engines.

For PM emission factors, the emission factor in autumn was significantly higher than that in summer. The main reasons for the difference were as follows. The background environments of summer and autumn were different. In terms of PM formation, the experimental results of existing studies were also verified. Temperature and humidity are known to be directly related to the formation of PM in the atmosphere [33]. In the case of relative road conditions, seasonal factors influence PM emissions, so PM emission factor estimation and emission inventories should be considered. The current study compared relevant worldwide research results on PM emission factors. The results are shown in

Table 3. Comparison of Emission Factors.

No.	Road Type, Location	Vehicle Type	PM2.5 Emission Factor mg/(veh.km)
1 [34]	Arterial road, London	LDV	10
		HDV	179
2 [35]	Highway, Copenhagen	LDV	11
		HDV	233
3 [36]	Tunnel, Brisbane	LDV	13
		HDV	124
4 [37]	Tunnel, Beijing	LDV	6.6 ± 2.5
		HDV	182.0 ± 149.3
	Tunnel, Tianjin	LDV	3.7 ± 1.5
		HDV	56.2 ± 58.4
	Tunnel, Tsingtao	LDV	3.5 ± 10.3
		HDV	133.4 ± 79.9
5 [32]	Tunnel, Mumbai	Fleet	44.2 + 26.9

. The LDV emission factor delivered by the current study approximated the estimations in China. This was probably the result of the uniform national vehicle emission standard, and that these cities are in developed coastal areas. The HDV emission factor was not only significantly higher than the LDV emission factor, it was also highly volatile, due to technical variation, aging, and vehicle emission inspection and maintenance (I/M).

Unlike gaseous emission modeling, vehicular PPM, especially diesel soot, has a complex formation process, with PM proceeding from a series of aggregation, coagulation, condensation, adsorption, and oxidation [31]. Afterward, PM may float or deposit depending on its diameter. The PM regression model hypothesized that the PPM mentioned above was the major source, while SPM was induced by tailpipe gaseous precursors. The regression models delivered emission factor estimations based on the PM formation mechanisms. Therefore, tailpipe PPM and its derivatives were modeled with traffic activities under regular meteorological conditions. The traffic facility-oriented PM study focused on traffic-induced PM, including PPM and SPM, to assess the adversary effects on the ambient environment. For normal days, the model estimated the traffic facility-oriented PM emissions via traffic volumes by varied vehicle categories.

This study delivered emission factors and corresponding standard deviations to gauge uncertainty. Apart from some minor issues, each set of coefficients was statistically significant. The models even considered the effect of temperature, with PM prone to be formed in autumn when the road sustained a similar volume of vehicle traffic. Prior research found a similar increasing PM mass concentration trend from summer to winter and indicated that emissions by the vehicle fleet contributed to the formation of SO₄²⁻ and NO₃⁻ [38]. Thus, the seasonal effect should be taken into consideration. Otherwise, the two PM datasets, namely, summer and autumn, were mixed, with the biased residuals suggesting that the model underestimated the emission factors in autumn. Emission factors for LDVs and HDVs were similar to other findings in Table 3. Moreover, the uncertainty of the HDV emission factor, as shown in the standard deviation, was higher. The high variation of PM emission factors was subject to technical variation, mileage, loading (bus and truck), fuel injection strategy, diesel/gasoline particulate filters, etc. [32].

4.2. Sampling Composition and PM Formation Analysis

The PM_2.5 samples were ultrasonically extracted by deionized water and filtered by disposable aqueous microporous membranes (pore size of 0.45 µm). Water-soluble inorganic ions (NH⁴⁺, NO₃^–, SO₄^2–, etc.) were measured by ion chromatography (ICS5000+, Thermo Scientific™ Dionex™, USA). The PM_2.5 sample solutions were exchanged and separated by the anion or cation chromatography column and detected by an inhibited conductivity detector. The qualitative analysis was conducted based on retention time, with the quantitative analysis conducted based on peak height or peak area. The field and laboratory blank filters were processed using the same methods and measured simultaneously.

The PM sampling data were collected at the roadside of the tunnel egress (exit) and at varied background points. The study also tried to conduct sampling during haze events. However, only one haze event occurred in Beijing after the Winter Olympic Games. In the end, two haze day samples and seven normal day samples were collected among all valid samples. In total, 11 anions and cations were obtained through sampling and chemical analysis. From them, seven categories of anions and cations were selected for the relatively high proportions needed for analysis.

To explore the mechanisms of PM formation during the haze event, the PM sample composition of inorganic ions was statistically analyzed to test whether it changed significantly under different meteorological conditions. The background value change was first tested. After analyzing some samples from the background site, the proportions of varied components were found, as shown in

Table 4. Portion of Water-Soluble Ions in Different Samples at Background Sites.

Date	Background PM Concentration	Li+	Na+	NH4+	K+	F−	NO3−	SO42−
15 May	46	0.0001	0.0214	0.1604	0.0115	0.0059	0.4791	0.3055
16 May	45	0.0001	0.0202	0.1756	0.0086	0.0016	0.4606	0.3121
21 May	8	0.0003	0.2718	0.0825	0.0638	0.0584	0.0746	0.3057
31 May	13	0.0005	0.0670	0.2226	0.0226	0.0376	0.1616	0.3988
9 April	90 *	0.0001	0.0153	0.1759	0.0104	0.0047	0.4489	0.3362

Notes: * Nearby meteorological stations (µg/m³); may not equal 1 due to rounding and omissions.

.

From the experimental data, the specific gravity of inorganic salt ions was found to be different under different weather conditions. When analyzing the differences between the samples with the Friedman test using the ordinals of anions and cations, if the air quality was good, the significant difference p-value of all samples was 0.084060, which was greater than 0.05. Hence, no significant difference was found in the collected PM samples when the weather quality was good. However, when haze event samples were analyzed with normal weather samples using statistical tests, the p-value was 0.000156, indicating a significant difference between the samples after adding the haze event data. Thus, the difference was caused by the haze event data.

The study then tested the PM samples collected from the roadside test under the haze condition, as shown in

Table 5. Sampling Concentrations of Water-Soluble Ions on Roadside.

Sampling Date	Background PM Concentration	Li+	Na+	NH4+	K+	F−	Cl−	NO3−	SO42−
15 May	46	0.002	0.615	3.631	0.117	0.060	1.017	11.958	4.323
23 May	55	0.002	2.944	1.648	0.281	0.105	2.223	12.124	3.052
25 May	54	0.002	1.551	2.482	0.127	0.096	2.037	8.062	3.765
10 April	96 *	0.002	1.985	1.005	0.393	0.169	3.382	43.329 **	12.352

Notes: * Nearby meteorological stations (µg/m³); ** ion concentration (mg/L).

.

On normal days, the average concentrations of nitrates (NO₃⁻) and sulfates (SO₄²⁻) were 10.71 mg/L and 1.88 mg/L, and their standard deviations were 3.71 mg/L and 0.52 mg/L, respectively. When the haze event occurred, nitrate and sulfate concentrations were 43.2 mg/L and 12.4 mg/L, respectively, thus showing a significant difference from those on normal days. Based on the composition analysis, NO₃^– and SO₄²⁻ concentrations dramatically increased during the haze event. As no major emission sources were in the vicinity, the nitrogen and sulfur contents were found to come from traffic emissions and other precursors transported over a long distance.

Compared to the data collected by aerosol mass spectrometers (AMSs) in Beijing during haze days, the sample composition varied significantly during the haze event. The shares of sulfate, and especially of nitrate, increased dramatically [19]. From the traffic data, however, vehicle operation speed varied by only a low magnitude. Therefore, the traffic facility-oriented PM needed to be considered more holistically. Remarkably, the haze event not only scaled up the total mass concentration, but a significant quantity of NO₃^– and SO₄²⁻ was also found at the background site during the haze event. As the major source of SO₂ is the combustion of fossil fuel [39], NO_x formation was the predominant contributor, mainly due to the Zeldovich mechanism in most combustion conditions. The formation of NO_x and PM showed a reversed relationship for diesel engines. The chamber temperatures can be decreased by using late injecting fuel inside the combustion chamber, increasing the soot formation rate [40]. Gas-phase NO₂ converts to particulate NO₃^– (p-NO₃^–) via photochemical reactions [41]. The primary sampling indicated NH₄⁺ concentration was low compared to those of NO₃^– and SO₄²⁻ on roadsides as the tunnel site lacked ammonia sources, such as agriculture and the open domestic wastewater of the Beijing metropolitan area [42].

Meanwhile, the sample composition during the haze event at background sites was statistically different from that of samples under normal meteorological conditions. Moreover, the sample collected within the tunnel had a similar composition profile to the background one. Due to the short lifetime of NO_x (typically less than 1 day), both local precursor transformation and regional precursor transportation attributed to the surge of PM mass concentration at transportation facilities, considering the conservation of nitrogen mass. As the formation mechanism changed significantly, SPM rather than PPM played a significant role. A disproportional increase of NO₃^– and SO₄²⁻ concentration occurred in the haze event. The gaseous NO_x and SO₂ were found to chemically convert into SPM within the tunnel during the haze event. The primary composition test suggested that the gaseous NO_x and SO₂ had significantly transformed into PM in the tunnel and at the background sites. However, the study could not exclude the reaction of non-traffic precursors within the tunnel. Therefore, the current PM emission model could not be established on haze event days. In terms of mass, the SPM was not correlated with traffic activities but with its formation rate. Thus, it was difficult to model traffic activities using a microscopic traffic simulation package.

5. Concluding Remarks

The framework for traffic facility-centered PM emissions was proposed based on data collected by traffic sensors and localized PM sensors. This study also delivered emission factors and corresponding estimation uncertainties by vehicle category in Beijing. Based on the emission factors collected over two seasons, the PM emission factors in summer are lower than the ones in autumn. And the PM emission factors of HDVs were significantly higher than those of LDVs. The standard deviations of emission factors suggest the technical conditions of the HDVs fleet are highly varied. Compared to gaseous pollutants, a state-of-the-art emission factor estimation relies on the PPM and SPM dynamics. Under normal meteorological conditions, the estimation of PM emission factors was found to correlate PPM and its derivatives with vehicle traffic activities. The emission model should also consider seasonal effects. It should have two sets of emission factors, one for summer and the other for winter, due to varied SPM formation resulting from temperature, humidity, and PBL height. The haze event was associated with dramatically increased PM mass concentration. According to sample composition analysis, chemical composition varied significantly during the haze event. Moreover, the primary composition test suggested that gaseous NO_x and SO₂ significantly transformed into PM within the tunnel. It suggested that SPM makes a critical contribution to the surge in mass concentration during haze events. However, the study could not exclude the reaction of non-traffic precursors in the tunnel. Therefore, the PM emission is difficult to model with traffic activities on haze days.

The study also has further implications for traffic policy regarding aerosol precursors, by diesel vehicles particularly. The PM_2.5 emission factor of diesel-powered HDVs was substantially higher than the LDV emission factor. Hence, the fleet management of diesel vehicles on haze days should be paid more attention to. Unilateral control of PPM induced by diesel vehicles is unlikely to be cost-effective. It may also be feasible to intervene with aerosol growth processes to reduce the total PM_2.5. During haze events, the mitigation of HDV traffic could reduce diesel vehicle PPM (soot) and NO_x, which are the SPM precursors. Therefore, the prevalence of particulate filters and reductions in aerosol precursor gases, NO_x and SO₂, will be essential traffic policies for PM control with the shifting focus from PM emissions in terms of mass concentration to PM number [43]. Other feasible policies also include but are not limited to further reducing the sulfur content of diesel fuels to reduce the synergy between sulfur and PM production and the emerging port charge compression ignition (PCCI) technology reducing the rates of both soot and NO_x [44].

In the future, further research will be conducted. This study has presented the primary results of PM sampling. Only one sample was collected at the roadside during the haze event, and more haze event samples should be gathered. In particular, more samples should be collected in long haze events that can last several days. The key component for traffic-induced emission estimation is SPM, with the chemical transformation mechanisms complex. Future work needs to explore the extent to which vehicle-emitted gaseous pollutants transform into SPM during haze events. In this study, the emission factor estimation was limited to a 60 km/h speed level. Further speed correction factors would be required.