Characterization of Dispersion and Ultrafine-particle Emission Factors Based on Near-roadway Monitoring Part I: Light Duty Vehicles

This paper presents near roadway light-duty vehicle (LDV) dispersion and ultrafine-particle (UFP) emission factors (EFs) as a function of vehicle flow rate, speed and mode of operation (free flow and congestion) using 300 5-min measurements of carbon dioxide (CO2), UFPs, meteorology and traffic conditions near a LDV only roadway. Measurements were made between 2014 and 2018 on 20 days for 2 to 3 hours per day. Near roadway LDV dispersion was estimated using a vehicle emission model (MOVES) for CO2 EFs, measured CO2 concentrations and vehicle flow rate. EFs for UFPs were estimated based on the calculated CO2 dispersion. This eliminated the need to rely on models to estimate dispersion. Results indicate that the near roadway LDV dispersion is a two-stage process. When there was a small number of vehicles (< 10,000 veh h–1), the vehicle induced dispersion was not continuous (unsteady-state conditions). Under unsteady-state conditions, the rate of LDV dispersion near the roadway increased from 2 to 6 m2 s–1 as the number of vehicles increased. For larger vehicle flow rates (> 10,000 veh h–1), the dispersion was continuous (steady-state continuous mixing). Under steady-state condition, the rate of dispersion was near 6 m2 s–1 and did not increased with additional vehicles. Dispersion models failed to response to unsteady-state conditions and predicted constant dispersion. The UFP EFs varied from 0.5 × 1013 to 1.5 × 1013 pt km–1 veh–1 for LDV (3x). The increase in UFP EFs was related to increase in vehicle flow rate and speed. The outcome of this study provides important EFs for urban planner to quantify air quality near roadway due to UFPs.


INTRODUCTION
Vehicle emission factors (EFs) that account for vehicle flow rate and type (light duty and heavy duty) and vehicle mode of operation (free flow and congestion) are needed to model vehicle emissions and provide insight into human exposure. This is due to the fact that heavy duty vehicles (HDVs) are usually under 5% of the total vehicle flow rate on highways but contribute more than half of the ultrafine particle (UFP) emissions (Abu-Allaban et al., 2002;Ntziachristos et al., 2007;Xiang et al., 2018). This makes the determination of EFs for HDVs very important. However, in order to estimate the real-world EFs, the dispersion process near roadways needs to be characterized. Researchers have determined that turbulent mixing is a dominate mechanism for UFP dispersion Wexler, 2002, 2004;. Thus, understanding the dynamic of dispersion is critical. However, there is no generally applicable method to directly relate dispersion to turbulence near roadways. Even though the enhancement in turbulence caused by HDVs has been studied widely, more research is needed that relates dispersion from HDV to turbulence from HDV.
The turbulence generated by vehicle motion is called vehicle-induced turbulence (VIT). VIT produces the dispersion of vehicle exhaust emissions in the near roadway environment and is related to mechanically induced circulations associated with the so-called "wake" behind vehicles. The region of flow immediately behind the vehicle is called the "wake". The wake is a complicated region of flow separations associated with moving vehicles that generate random, three-dimensional, flow fluctuations (turbulence).
VIT has been evaluated using wind tunnels, computational fluid dynamic models (CFD) and field measurements. Wind tunnel studies (Eskridge and Thompson, 1982;Pavageau and Schatzmann, 1999;Kastner-Klein et al., 2000;Carpentieri et al., 2012;McArthur et al., 2016;Lo and Kontis, 2017) have demonstrated that the extent of the wake depends on the height of the vehicle, the aspect ratio (length/height), type of vehicle and the vehicle mode of operation (free flow and congestion). The largest wake behind light duty vehicles (LDVs) extends to about 15 times the vehicle height (Eskridge and Hunt, 1979;Eskridge and Thompson, 1982) and HDVs produce significantly larger wakes than LDVs (McArthur et al., 2016;Lo and Kontis, 2017;Miller et al., 2019). The results from these studies indicate that variation in VIT cannot be neglected when evaluating dispersion in near roadway environments.
VIT has also been studied using CFD models that simulate dispersion near roadways as well as vehicle flow patterns to estimate pollutant concentrations (Sahlodin et al., 2007;Wang et al., 2009;Wang and Zhang, 2012;Wang et al., 2013;Huang et al., 2014;Bhautmage and Gokhale, 2016;Kim et al., 2016). These results were then compared to measured concentrations to evaluate modeled turbulence and dispersion. In computational models, dispersion is usually considered to be an empirical property of the turbulent flow. Results from CFD studies indicate that by assuming VIT is uniform within the wake structure, the VIT from vehicle movements can eventually reach a steady-state where the continuity of the vehicle flow prevents the turbulence from decaying. Under these conditions, the flow rate of the vehicles is large enough to produce a continuous, steady-state flow where the VIT does not increase with an increase in vehicle flow rate. When vehicle flow rate is low, the VIT is not continuous and increases with vehicle flow rate. This represents unsteadystate flow.
Field experiments have been conducted to assess VIT near roadways from vehicles (Trivikrama Rao et al., 1979;Chock, 1980;Rao et al., 2002;Kalthoff et al., 2005;Alonso-Estébanez et al., 2012;Gordon et al., 2012;Tong et al., 2015;Miller et al., 2019). These field studies indicate that there is a significant increase in turbulence near roadways due to vehicle movements on roadways and VIT can be affected by vehicle type, flow rate and mode of operation.
In this study, near roadway dispersion was estimated based on modeled CO 2 EFs and measured vehicle flow rate and CO 2 concentrations. 664 samples were collected at 5-min interval on a total of 60 days for a 2 to 3 hour sampling period between 2015 and 2018. CO 2 emission factors (EFs) were calculated from MOVES for different vehicle type, flow rate and mode of operation (free flow and congestion) and combined with CO 2 and traffic measurements to calculate real-world dispersion near a roadway from LDVs and HDVs. Variations in dispersion of CO 2 were determined as a function of vehicle flow rate and vehicle mode of operation. Since the vehicle fleet is usually a mixture of LDVs and HDVs, it is difficult to accurately separate the UFP contribution due to LDVs and HDVs. To address this issue, UFP EFs of LDVs were determined based on a previous study at an LDV-only roadway (Xiang et al., 2019, this issue) and used to calculate EFs from HDVs.

Monitoring Site
Field measurements were conducted near the Dan Ryan Expressway (DRE) in Chicago, IL a few kilometers south of the city center (see Fig. 1) and next to the Illinois Institute of Technology campus. Information from the Illinois Department of Transportation (IDOT, 2016) provided the annual average daily traffic flow rate of 300,000 veh day -1 with 8.4% HDVs. The sampling program was conducted during the summer and fall from 2015 to 2018 (total of 664 5-min measurements). Each 5-min was a separate sampling period. The freeway is depressed 6 meters and located between a parking lot and low-rise buildings. The freeway has 14 lanes, separated into north and south bound roadways by a 10-m wide subway platform. For each direction, four express lanes are restricted to LDVs and three local lanes are open to both LDVs and HDVs. CO 2 , UFP and meteorology measurements were collected with handheld, fast-response instruments simultaneously at a near roadway and a background location. Near roadway sampling was conducted on the east side of the subway platform (site A) 3 meters from the edge of the roadway at an elevation of 4 meters when there were easterly winds. Background measurements were recorded upwind, more than 200 m away from the roadway (not shown). The traffic information was recorded on an overpass (33 rd street) that allows all lanes of traffic to be (flow rate and speed) recorded. The determination of vehicle types was made manually, LDVs were defined as passenger cars and trucks while HDVs were defined as long combination haul with either 10 or 18 wheeled (Gordon et al., 2012). Since site A was in the middle of DRE, only oneside traffic (northbound) were used to evaluate near roadway measurements. Details concerning the sampling program are reported elsewhere (Xiang et al., 2018). Sampling was conducted only when easterly winds occurred, and wind direction was within 45° of normal to the roadway. Only data with all lanes free flow or all lanes congestion was used for data analysis.

Estimation of near Roadway Dispersion
The near roadway dispersion was calculated based on theoretical relationships between vehicle emissions, meteorology and pollutant concentrations (Palmgren et al., 1999). To characterize near roadway dispersion, a tracer gas with a well-known emission factor (EF) is needed (Imhof et al., 2005;Zhang et al., 2005;Jones and Harrison, 2006;Wang et al., 2010;Nickel et al., 2013). In this study, CO 2 was chosen as the tracer gas. The reason that CO 2 is an ideal tracer gas was demonstrated in a previous study (Xiang et al., 2019, this issue). Near roadway fleet dispersion can be estimated with EFs of CO 2 , vehicle flow rates and CO 2 concentration as shown in Eq. (1).
In Eq. (1), D fleet is near roadway fleet dispersion (m 2 s -1 ). EF CO2,LDV and EF CO2,HDV are emission factor of CO 2 (g m -1 veh -1 ) modeled by MOVES for LDVs and HDVs, respectively. N LDV and N HDV are the vehicle flow rate of the LDVs and HDVs, respectively (veh s -1 ). The differences between near roadway concentrations (C CO2 ) and background concentrations (C CO2,BG ) provides the CO 2 concentrations contributed by vehicle emissions (g m -3 ). The 5-min near roadway measurements are subdivided into (1) all lanes free flow and all lanes congestion conditions, (2) uniform fleet vehicle flow rate intervals of 1,000 veh h -1 . Categorizing the near roadway fleet dispersion by vehicle flow rate and mode of operation allows the estimation of the vehicle fleet EFs as a function of vehicle flow rate, speed, mode of operation and type of vehicle (HDVs and LDVs). In this study, for each vehicle type (LDVs and HDVs) and mode of operation (free flow and congestion), D fleet is assumed to occur in the vehicle wake and considered as a combination of dispersion from both LDVs and HDVs (D fleet = D LDV + D HDV ). This linear relationship was suggested by other researchers and assume no interaction between wakes from different vehicle (Benson, 1984;Grønskei, 1988;Berkowicz et al., 1997).
The ensemble means are calculated for each vehicle flow intervals (1,000 veh h -1 ) and used to evaluate dispersion. The use of ensemble means allow changes in measured pollutant concentrations due to variations in vehicle flow rate and dispersion to be evaluated. (Arya, 1999;Seinfeld and Pandis, 2016).

Determination of UFP EFs for Traffic Fleet
Since the CO 2 and particles are homogeneously mixed (Kurniawan and Schmidt-Ott, 2006;Ntziachristos et al., 2007;Rönkkö et al., 2017;Hietikko et al., 2018), UFP emission factors (EFs) can be estimated using D fleet , measured UFP concentrations and vehicle flow rates (See Eq. (2)). This approach eliminates the need to rely on a dispersion model to determine the near roadway dispersion.
In Eq. (2), EF UFP,fleet is the vehicle fleet UFP emission factor (pt m -1 veh -1 ). C UFP and C UFP,BG are UFP concentrations measured at near roadway and background locations, respectively (pt m -3 ). Differences between C UFP and C UFP,BG are considered as UFP contributed by vehicles. N fleet is total vehicle flow rate (veh s -1 ), that is the sum of the number of LDVs and HDVs (N fleet = N LDV + N HDV ). Because UFP and CO 2 were simultaneously measured, total vehicle flow rate (N fleet ) and near roadway fleet dispersion (D fleet ) from Eq. (1) can be used in Eq. (2).
One of the critical problems when analyzing roadway data is that the vehicle fleet consists of both LDVs and HDVs. To differentiate the EFs for different vehicle types (LDVs and HDVs), a mass balance equation, Eq. (3) can be applied. A similar equation has been used by other researchers (Palmgren et al., 1999;Imhof et al., 2005;Morawska et al., 2005;Zhang et al., 2005;Wang et al., 2010;Nickel et al., 2013;Krecl et al., 2018). LDV EFs determined for Lake Shore Drive (LSD) are available as input to solve Eq. (3) for EFs for HDVs (Xiang et al., 2019).
In Eq. (3), EF UFP,LDV and EF UFP,HDV are the UFP emission factor for LDV and HDV, respectively (pt m -1 veh -1 ). N LDV and N HDV are the vehicle flow rate for LDVs and HDVs, respectively (veh s -1 ). Eq. (3) was used for each data interval. By solving Eq. (3), UFP EFs for HDVs can be estimated as a function of HDV flow rates and vehicle mode of operation.

Overview of Meteorological Condition
In this study, the average temperature was 22 (± 3)°C for free flow and 21 (± 4)°C for congestion; the average relative humidity was 53 (± 12)% for free flow and 60 (± 12)% for congestion. Atmospheric stability class were neutral or stable (class D) for most of the sampling days. On the average, free flow had higher wind speeds (2.0 ± 0.7 m s -1 ) than congestion (1.5 ± 0.4 m s -1 ). Fig. 2 provides information on the vehicle speed-flow relationship during the sampling program and is typical of traffic conditions on many roadways (Mannering, 2013). The figure indicates that there were significant variations in vehicle driving conditions that can be related to vehicle EFs. The figure indicates that vehicle flow rate is at a maximum capacity at a speed near 70 km h -1 for LDVs and 55 km h -1 for HDVs and decreases for higher and lower speeds. The driving condition is considered free flow on the upper part of the speed-flow curves which represents conditions when the vehicles are unhindered by other vehicles. Under this driving condition, as the vehicle flow rate increases the distance between individual vehicles decreases and results in a reduction of the vehicle speed. The average speed of LDVs for free flow conditions was 110 km h -1 and 28 km h -1 for congestion. For free flow conditions, HDV speed ranged from 60 to 110 km h -1 with an average near 90 km h -1 . For congestion, the average speed for HDV was 26 km h -1 with a range from 9 to 50 km h -1 . The express lanes generally had a higher speed than the local lanes and HDVs generally had a lower speed than LDVs. The large variations in speed and flow for vehicles indicated that there was significant variation in traffic conditions. These variations were able to be observed due to the short sampling time (5-min). For each 5-min sample, total vehicle flow rate (N fleet ) is equal to sum of LDV in express and local lanes (N LDV ) and HDV in local lanes (N HDV ). The range of the total vehicle flow rate is between 5,000 and 10,000 veh h -1 . Fig. 3 presents the CO 2 EFs for LDVs and HDVs determined using MOVES as a function of vehicle flow rate under different traffic conditions (free flow and congestion). The LDV EFs were near 200 g km -1 veh -1 and the HDV EFs were near 1,200 g km -1 veh -1 for free flow conditions. For congestion conditions, both the LDV and HDV EFs increased significantly (75% for LDVs and 100% for HDVs) when the total vehicle flow rate decreased from 10,000 to 3,000 veh h -1 . For the overall average, HDV CO 2 EFs are 6 times higher than LDVs for free flow and congestion conditions. The trend of CO 2 EFs in this study is in accordance with other studies. Liu and Frey (2015) reported that LDV CO 2 EFs increased from 190 to 300 g km -1 veh -1 as the speed decreased from 110 to 16 km h -1 . Zhang et al. (2011) reported HDV CO 2 EFs varied from 1,030 to 1,700 g km -1 veh -1 when speed decreased from 100 to 30 km h -1 .

CO2 and UFP Concentrations near Roadway
Variations in the near roadway CO 2 and UFP measured concentrations (background subtracted) as a function of total vehicle flow rate are shown in Fig. 4. The data for each HDV category were fitted by polynomial regression to represent the vehicle flow rate. For free flow and congestion conditions,  The 5-min data were first divided into 1,000 veh h -1 intervals then categorized into different HDV flow rates. Emission factors for each interval were modeled using MOVES, which requires traffic speed, temperature and relative humidity. MOVES default driving cycles and age distribution were used. Bars indicate one standard deviation.
concentrations of both CO 2 and UFP decreased with an increase in fleet flow rate. The minimum concentration occurred at the highest vehicle flow rate. Application of Eq. (1) (next section) indicates that there is a significant increase in near roadway dispersion when there is an increase in the number of HDVs and this accounts for the decrease in concentrations at high flow rates. In a previous study near a LDV-only roadway (Xiang et al., 2019), concentrations of both CO 2 and UFP increased when total vehicle flow rate increased (See Fig. 4(b)). These results indicate that HDVs have a significant impact on dispersion of pollutant concentrations near roadways because one would expect the concentration to increase with vehicle flow rate.

Variation in Near Roadway Dispersion with Variation in HDV Flow Rate
Eq.
(1) was used to calculate near roadway fleet dispersion (D fleet ) for each 1,000 veh h -1 interval. The calculations are based on the modeled CO 2 EFs in Fig. 3 and the ensemble mean CO 2 as a function of total vehicle flow rate and HDV flow rate from Fig. 4(a). Example calculations of dispersion as a function of vehicle flow rate and type are shown in Table 1. Table 1 indicates that the ratio of HDV/LDV dispersion varied between 25 and 40 depending on traffic conditions (HDV flow rate and speed). On the average, near roadway dispersion from one HDV is, nearly 30 times higher than one LDV. Even when the vehicle fleet flow rate was similar (9,000 to 10,000 veh h -1 ), increasd HDV flow rate resulted in an increase in D fleet . This is because of the differences in the ratio of HDV/LDV dispersion. Fig. 5 presents the variations of D fleet as a function of total flow rate for different HDV flow rates near DRE. Polynomial least-square regression was used to fit the data. D fleet increased when both the total vehicle flow rate and the HDV flow rate increased for free flow and congestion conditions. However, the largest increases are associated with increases in HDV flow rate. D fleet increased nearly 3 times (6-18 m 2 s -1 ) at a fleet flow rate of 10,000 veh h -1 when the HDV flow rate increased from 0 to 1,000 veh h -1 . When the information in Figs. 2 and 5 are combined, one can see that D fleet varies with both vehicle speed and vehicle flow rate. Compared to the study conducted near a LDV-only roadway (Xiang et al., 2019), one unit of increase in vehicle fleet flow rate (1,000 veh h -1 ) in DRE produce more D fleet depending on the HDV flow rate in vehicle fleet. This is due to the fact that one HDV can produce more dispersion than one LDV, as indicated in Table 1. Fig. 6 provides the calculated fleet UFP EFs as a function of total vehicle flow rate, HDV flow rate, and vehicle mode of operation using Eq. (2). The vehicle fleet EFs for free flow conditions decreased when total vehicle flow rate increased from 5,000 to 10,000 veh h -1 with an average value of 0.52 × 10 14 (±0.17 × 10 14 ) pt km -1 veh -1 . However, the vehicle fleet EFs increased by 4 times when the HDV flow rate increased from 0-200 to 800-1000 veh h -1 . For congestion conditions, average fleet EFs were 0.87 × 10 14 (±0.18 × 10 14 ) pt km -1 veh -1 , and when the vehicle flow rates decreased fleet EFs increased rapidly (total increase is 6 times). These results can be compared to UFP EFs for LDVs that increased by 20% (from 0.1 × 10 14 to 0.12 × 10 14 pt km -1 veh -1 ) with LDV flow rate from 5,000 to 10,000 veh h -1 (Xiang et al., 2019). The 5-min data were first divided into 1,000 veh h -1 intervals then categorized into different HDV flow rates. LSD data for UFP concentrations were modified from Xiang et al. (2019). Bars indicate one standard deviation. Fig. 7 provides the estimated HDV UFP EFs as a function of HDV flow rate, speed, and mode of operation using Eq. (3). HDV EFs were calculated based on total vehicle fleet EFs in Fig. 6 and the LDV EFs provided from a previous study (Xiang et al., 2019). HDV EFs were 6.6 × 10 14 (±1.2 × 10 14 ) pt km -1 veh -1 for free flow and were 2 times higher for congestion condition (16 ± 3.7 × 10 14 pt km -1 veh -1 ). To better understand the dynamic of HDV UFP EFs as a function of vehicle mode of operation, Fig. 7 also presents HDV UFP EFs as a function of HDV speed. As the HDV speed decreased from 100 to 10 km h -1 , HDV UFP EFs increased 3 times.

CONCLUSIONS and DISCUSSION
Variations of near roadway fleet dispersion (D fleet ) were calculated using 5-min CO 2 measurements, vehicle flow rate # D LDV is calculated using the best fit line for LSD data from previous study (Xiang et al., 2019). $ D HDV is calculated using D LDV from LSD, where D HDV = D fleet -D LDV . ^ LDV normalized dispersion is defined as near roadway LDV dispersion (D LDV ) divided by LDV flow rate (N LDV ). & HDV normalized dispersion is defined as near roadway HDV dispersion (D HDV ) divided by HDV flow rate (N HDV ). * Ratio of HDV/LDV dispersion is the ratio of HDV normalized dispersion to LDV normalized dispersion.  (1). The data were first divided into 1,000 veh h -1 intervals then categorized into different HDV flow rates (See Table 1). LSD data is modified from Xiang et al., 2019. Bars indicate one standard deviation.
and a vehicle emission model (MOVES). Results indicate that HDVs not only generate significantly more UFP emissions than LDVs they also produce significantly more dispersion. The average ratio of HDV to LDV dispersion from individual vehicles was near 30 for free flow conditions. Based on Eq. (1) and Fig. 5, the near roadway fleet dispersion increased when the flow rate of HDVs increased under both free flow and congestion conditions. One of the important conclusions of this study is that the enhancement of dispersion due to HDVs leads to enhanced dispersion for both LDVs and HDVs emissions. This resulted in the lowest UFP and CO 2 pollutant concentrations occurring during the highest vehicle flow rates. Identifying the near roadway dispersion from HDVs is critical because the HDVs provide the dispersion of emissions from both HDVs and LDVs. This study also provides a comprehensive set of UFP emission factors (EFs) for HDVs as a function of vehicle flow rate, speed, and mode of operation. The average ratio of HDV to LDV UFP EFs was 44 for free flow and 125 for congestion. The emissions from individual HDVs were 3 times higher during congestion than free flow conditions. HDVs EFs ranged from 10 14 to 10 15 pt km −1 veh −1 and LDVs varied from 10 13 to 10 14 pt km −1 veh −1 depending on vehicle flow rate and mode of operation. Results from other studies are presented in Table 2 (Imhof et al., 2005;Morawska et al., 2005;Birmili et al., 2009;Wang et al., 2010;Nickel et al., 2013;Krecl et al., 2017;Krecl et al., 2018). For free flow condition (HDV speed > 55 km h −1 ), EFs ranged from  7 × 10 14 to 30 × 10 14 pt km −1 veh −1 . For congestion conditions, the range was from 1 × 10 14 to 55 × 10 14 pt km −1 veh −1 . UFP EFs from this study are within the range of the average value reported from other studies (See Fig. 8). The wide range of reported UFP EFs reported in the literature indicates the need to evaluate addition information to address the differences in (vehicle operation, instrumentation, road type and sample site location).
When compared to the study that were conducted near a LDV-only roadway, the UFP EFs for HDVs not only show significant larger values but also present a different trend as a function of vehicle speed (Xiang et al., 2019, this issue).  This can be caused by different operating mechanism in gasoline and diesel engines and distinct driving cycles and this requires that need more real-world and laboratory investigations.

ACKNOWLEDGMENTS
We would like to thank all participants in the field measurement, in particular Liuyang Sun, Xuexuan Peng for collecting field samples.

DISCLAIMER
Reference to any companies or specific commercial products does not constitute endorsement.