Near-road Vehicle Emissions Air Quality Monitoring for Exposure Modeling

Abstract

Exposure to vehicular emissions has been linked to numerous adverse health effects. In response to the arising concerns, near-road monitoring is conducted to better characterize the impact of mobile source emissions on air quality and exposure in the near-road environment. An intensive measurement campaign measured traffic-related air pollutants (TRAPs) and related data (e.g., meteorology, traffic, regional air pollutant levels) in Atlanta, along one of the busiest highway corridors in the US. Given the complexity of the near-road environment, the study aimed to compare two near-road monitors, located in close proximity to each other, to assess how observed similarities and differences between measurements at these two sites inform the siting of other near-road monitoring stations. TRAP measurements, including carbon monoxide (CO) and nitrogen dioxide (NO₂), are analyzed at two roadside monitors in Atlanta, GA located within 325m of each other. Both meteorological and traffic conditions were monitored to assess the temporal impact of these factors on traffic-related pollutant concentrations. The meteorological factors drove the diurnal variability of primary pollutant concentration more than traffic count. In spite of their proximity, while the CO and NO₂ concentrations were correlated with similar diurnal variations, pollutant concentrations at the two closely sited monitors differed, likely due to the differences in the siting characteristics reducing the dispersion of the primary emissions out of the near-road environment. Overall, the near-road TRAP concentrations at all sites were not as elevated as seen in prior studies, supporting that decreased vehicle emissions have led to significant reductions in TRAP levels, even along major interstates. Further, the differences in the observed levels show that use of single near-road observations will not capture pollutant levels representative of the local near-road environment and that additional approaches (e.g., air quality models) are needed to characterize exposures.

Introduction

Measurement of traffic-related air pollutants (TRAPs) in near-road environments has been an active area of research given the extensive evidence linking exposure to primary traffic pollution to a range of adverse health effects (Brugge et al. 2007; Chen et al. 2017; Golan et al. 2018; HEI 2009a, b; Liang et al. 2019; Smith et al. 2017; WHO 2005). High levels of air pollutants exist within near-road microenvironments due to vehicle exhaust and mechanically-generated emissions (Baldauf et al. 2013; Karner et al. 2010). Concentrations of TRAPs, including nitrogen oxides (NO_x), volatile organic compounds (VOCs), carbon monoxide (CO), primary fine particulate matter (PM_2.5), black carbon (BC), and organic carbon (OC), in particular, have been found to be elevated near heavily trafficked roads (Baldauf et al. 2012; Beckerman et al. 2008; Boogaard et al. 2011). Such emissions can interact chemically and physically with each other and other pre-existing pollutants in the roadway environment, leading to a complex, multicomponent mixture (Saha et al. 2018).

Historically nitrogen dioxide (NO₂) concentrations exceeded the hourly National Ambient Air Quality Standard (NAAQS) in the near-road environment (Vijayaraghavan et al. 2014), though emissions controls have led to reduced pollutant concentrations and exceedances of the national standards (DeWinter et al. 2018). Where high NO₂ measurements are still a concern, vehicle emissions are the major contributing source (Lin et al. 2016). Elevated TRAP concentrations are affected by a number of factors including traffic volume, vehicle types, local meteorological conditions (Karner et al. 2010), and local topography such as natural (Baldauf 2017) and built highway features (Baldauf et al. 2016). These conditions can result in a wide range of pollutant concentrations observed along different roadway segments within the same urban environment (McAdam et al. 2011). A concern for the possibility of high concentrations of NO₂ initiated the U.S. Environmental Protection Agency (EPA) to implement a national near-road monitoring network to specifically measure TRAPs (EPA 2010, 2011). The near-road monitoring network focuses on locating monitors near the most heavily trafficked roads in urban cores around the country. One objective for the network was to improve exposure assessments to primary traffic emission for urban populations vulnerable to this pollution source. The implementation of the monitors began January 1, 2014 and as of January 1, 2015, there were 61 active monitoring sites, two of which were located within 10m of two different heavily trafficked highways in Atlanta, GA.

While the near-road remains a potential high exposure environment, the near-road environment has changed with improved vehicle engine technologies, the associated emissions control systems, and fuel regulations leading to reduced mobile source emissions affecting near-road concentrations (Ayala et al. 2012; Henneman et al. 2015; Karner et al. 2010; McDonald et al. 2015; Vijayaraghavan et al. 2014). A decrease in mobile emissions in Georgia over the last decade has contributed to an estimated 30% reduction in PM from mobile sources (Zhai et al. 2017). New studies are needed to revisit potential changes in the local spatial and temporal patterns.

The Dorm Room Inhalation to Vehicle Emissions (DRIVE) study was conducted in Atlanta, GA in 2014 along one of the busiest highway corridors in the US to characterize factors leading to human exposures in the near-road environment, both indoors and outside, and to assess statistical and integrated traffic exposure metrics for applications in epidemiological studies (Liang et al. 2018a; Liang et al. 2018b; Sarnat et al. 2018). Utilizing measurements from the DRIVE study, this paper aims to characterize pollutant dynamics in the changing near-road environment and to examine how well the near-road network monitors characterize TRAP concentrations in complex, dynamic urban environments. This issue is addressed here by focusing on the dynamics of TRAP observations in relationship to meteorology, traffic characteristics, and regional air pollution mixtures using two near-road monitoring sites in close proximity to each other along the same highway segment located near downtown Atlanta. Understanding how well a single monitor represents the concentrations across an urban area has importance in both regulatory as well as health assessment frameworks. Therefore, as part of this analysis we consider how the characterization of exposure and potential NAAQS exceedances might be different at two near-road monitors along the same road segment.

Methods

Concentrations of primary tailpipe pollutants are used as tracers for the impact of on-road mobile emissions to urban environments. Measurements were conducted continuously along a major highway in Atlanta, GA at two near-road highway monitors located in relatively close proximity to each other (325 m). One of the near-road locations was part of the DRIVE study and the second was part of the national EPA Near-road Monitoring Network operated by the Georgia Environmental Protection Division (GA EPD) of the Department of Natural Resources since June 2014. This analysis assesses hourly concentration measurements from two near-road monitoring locations along the same highway segment, and examines the impact of key meteorological and traffic factors on corresponding measured pollutant concentrations. Through comparing concentrations at the two sites within 325m of each other, the analysis can further assess how representative the EPA near-road network monitors are for understanding microscale exposure to populations within the near-road environment of an urban area.

Site descriptions

This study domain was centered around a segment of arterial interstate where Interstate 75 and Interstate 85 (I-75/I-85) merge in the center of Atlanta, Georgia (Fig. 1). In 2014, this highway segment along which the two monitoring locations are situated had an average annual daily traffic (AADT) of 330,000, composed primarily of light-duty gasoline passenger cars and trucks. Heavy-duty diesel trucks made up approximately four percent of the average daily vehicles on this portion of the highway (GA DOT 2012). Surface streets to the east of the highway follow a gridded pattern with an average block length of 450 feet and an AADT of more than 15 times less than the AADT for the highway segment. The land west of the highway segment for 1.5 km is the Georgia Institute of Technology (GIT) campus with limited vehicle access and much lower AADT. Within the study domain area, the Southeastern Aerosol Research and Characterization (SEARCH) network has maintained an urban background (UB) monitoring site since 1998 (Hansen et al. 2003). The site is located 2.3km west of the highway and is a long-term dataset that represents the historical Atlanta background concentration.

Figure 1.

(a) Map of sampling area. NR GIT site part of the EPA Near-road Monitoring Network in metro Atlanta, UB site part of the SEARCH Network. (b) Wind rose of hourly observations at the NR DRIVE site from September 8, 2014 to January 5, 2015

The near-road DRIVE sampling location (NR DRIVE) was located about a meter from the west side of the fifteen-lane highway (eight southbound and seven northbound) to the south of 10^th Street and to the north of North Avenue (Fig. S1). The monitoring site was located in a parking lot with less than 85 passenger vehicle spots and the vertical height from the highway to the parking lot was 0.5 m. The nearby EPA near-road monitoring network site (Fig. S2) was located on the Georgia Institute of Technology campus (NR GIT) 325m north of the NR DRIVE site location and about a meter from the west side of the highway. Trees were removed from the vegetation barrier to provide space for the site and a small, gated parking lot for about 100 passenger vehicles is to the west of the site.

Air Quality Instrumentation, Meteorological Characterization, and Traffic Data

Air quality monitors at the NR DRIVE location collected continuous ambient air samples from September 8, 2014 to January 5, 2015. All species concentrations, meteorological parameters, and traffic data were measured at local standard time (LST); however, daylight savings time did end on November 2, 2014. Continuous measurements of black carbon (Magee Aethalometer AE31), carbon monoxide (Thermo Model 48i), ozone (Thermo Model 49C), and nitrogen oxides (Teledyne API 200A) provided concentration data for pollutants commonly associated with vehicle emissions. Real-time gas analyzers collected measurements at 5-second averaging periods and the real-time black carbon monitor collected at 2-minute intervals. Data was collected using DAQFactory and WinWedge Pro software. Multipoint calibrations, zero air, and span checks provided an assessment for accuracy throughout the study and were used in time-weighted adjustments to the data. The sampling inlet height was approximately 3m and was 7m from the closest highway lane. All continuous data were averaged to hourly levels to assess temporal variability differences between pollutants and possible indicators. Details of the instrumentation and quality assurance can be found elsewhere (Sarnat et al. 2018). Continuous CO and NO_x data from the NR GIT site began on July 1, 2014. The sampling inlet height was approximately 3m and was 6m from the closest lane. The hourly concentration data were downloaded from the EPA air quality system (AQS).

Traffic vehicle count and speed data were obtained from the Georgia Department of Transportation Office of Transportation Data (GDOT 2014). The vehicle count data were collected at a location on I-75/I-85 1.5 miles south of the measurement location using Automatic Traffic Records. No major on or off ramps are located between the traffic count location and the near-road monitoring locations. Meteorological data collected at the NR DRIVE site (HOBO U30, Onset Corp) included wind speed, wind direction, temperature, and relative humidity. Wind speed and direction measurements used a cup anemometer and wind vane sensor. The sensors were mounted to a pole 5m above the ground with the temperature/relative humidity sensor placed in a protective solar radiation shield. The wind rose (Figure 1) from the NR DRIVE location found winds from the east (between 45 to 135 degrees) 77% of the time during the study period. With the NR DRIVE and NR GIT sampling locations west of the highway, winds from the east lead to downwind concentrations measurements. Atmospheric mixing height can be a critical factor for pollutant dispersion. For this reason, mixing height data was modeled using the Weather Research and Forecast (WRF) model (NCAR) for the 4km grid that included the study domain.

Multivariate regression modeling

We used multivariate linear mixed regression modelling to assess the factors that affected the temporal variability in the concentration of each TRAP:

$$P_t=\beta Z_t+{\theta }_t+{\epsilon }_t$$ (Eq. 2)

where P_t denotes the concentration of BC, CO, NO, NO₂, NO_x, or O₃ measured during hour t and β is the coefficient of interest that describes the influence of factor Z_t on the hourly pollutant level. The factors assessed include time period of the day (categorical), temperature (continuous), wind speed (continuous), relative humidity (continuous), wind direction (categorical), weekend (categorical, Saturday and Sunday), and hourly traffic counts (continuous). The temporal factor was divided into four periods: morning rush hour (6 – 9am), mid-day (10am – 3pm), evening rush hour (4 – 8pm, used as reference group), late evening (9 – 24pm), and early morning (1 – 5am). The wind direction factor was divided into three directions: north (315 – 45 degrees), east (45 – 135 degrees, which leads to the monitoring sites being downwind of the highway), and south (135 – 225 degrees). θ_t represents time-specific random intercepts used to capture potential variations not explained by Z_t and ε_t represents residual random normal error. The regression relationship between pollutant concentrations and driving factors developed a simplified method compared to the use of chemical transport models or dispersion models. A positive regression coefficient indicates an association between the pollutant concentration level at the measurement site and the factor, while controlling for all other factors included in the model. The multivariate regressions provide a direct relationship for health studies to better understand the driving factors for near-road exposures.

Results and Discussion

Observed near-road air pollutant concentrations

The NR DRIVE site and the NR GIT site measured CO, NO, NO₂, NO_x, and BC concentration continuously from September 8, 2014 to January 5, 2015. The NR GIT site began sampling BC from November 3, 2014. The NR DRIVE site also measured ozone (O₃), wind speed, and temperature. CO and NO₂ mean (standard deviation) concentrations at the NR DRIVE site were 425 ppb (210 ppb) and 29 ppb (15.5 ppb), respectively. At the NR GIT site, average (standard deviation) concentrations measured for CO and NO₂ were 624 ppb (338 ppb) and 19.5 ppb (8.6 ppb), respectively (Table 1 and Table 2).

Table 1.

Hourly averages of NR DRIVE and NR GIT near-road continuous instrumentation. September 8, 2014 to January 5, 2015.

		BC (ug m−3)		CO (ppb)		NO (ppb)		NO2 (ppb)		NOx (ppb)
		DRIVE	GIT	DRIVE	GIT	DRIVE	GIT	DRIVE	GIT	DRIVE	GIT
	N	2282	1115	2178	2816	2666	2798	2666	2798	2666	2798
Total	Mean	1.6	1.7	425	624	21	38	29	20	50	57
9/8–1/5	SD	1.3	1.2	210	338	24	29	16	8.6	35	34
	IQR	0.7 – 2.2	0.86 – 2.2	278 – 515	400 – 800	5.6 – 29	17 – 51	17 – 38	13 – 25	25 – 67	32 – 74
	Min-Max	0.06 – 13	0.13 – 9.4	133 – 1860	0 – 2200	0 – 201	1.0 – 233	2.4 – 94	2.3 – 52	2.7 – 252	4.9 – 263
	N	469	-	526	543	419	540	419	540	419	540
September	Mean	1.9	-	414	689	18	34	32	19	50	52
9/1–9/30	SD	1.2	-	157	264	15	20	13	6.4	26	24
	IQR	0.9 – 2.5	-	291 – 502	500 – 900	6.9 – 28	18 – 45	22 – 39	14 – 23	30 – 66	34 – 68
	Min-Max	0.2 – 8.8	-	136 – 1306	0 – 1500	0 – 95	1.3 – 118	4.4 – 82	4.8 – 42	3.7 – 174	11 – 139
	N	529	-	672	713	663	720	663	720	663	720
October	Mean	1.7	-	404	584	20	31	33	20	53	50
10/1–10/31	SD	1.4	-	185	242	21	23	17	8.1	33	27
	IQR	0.7 – 2.3	-	275 – 498	400 – 700	5.9 – 26	14 – 42	19 – 44	13 – 25	27 – 71	30 – 66
	Min-Max	0.2 – 12	-	133 – 1304	0 – 1700	0.2 – 155	1.0 – 125	2.8 – 94	2.3 – 49	3.7 – 202	5.8 – 150
	N	552	251	662	707	720	700	720	700	720	700
November	Mean	1.4	1.4	430	665	20	38	29	21	49	59
11/1–11/30	SD	1.3	1.1	227	356	25	33	17	11	39	40
	IQR	0.6 – 1.9	0.7 – 1.8	270 – 524	400 – 800	4.9 – 24	14 – 51	15 – 39	12 – 28	21 – 62	28 – 77
	Min-Max	0.2 – 13	0.2 – 5.8	138 – 1594	100 – 2100	0 – 180	1.0 – 186	4.2 – 82	2.8 – 52	4.0 – 225	4.9 – 213
	N	612	744	318	733	744	722	744	722	744	722
December	Mean	1.6	1.9	477	677	24	49	26	19	50	68
12/1 – 12/31	SD	1.4	1.3	280	367	28	35	13	8	37	40
	IQR	0.7 – 2.0	1.0 – 2.4	285 – 593	500 – 900	5.5 – 34	23 – 65	16 – 34	12 – 24	24 – 67	39 – 89
	Min-Max	0.06 – 12	0.1 – 9.4	168 – 1860	0 – 2200	0.05 – 201	1 – 233	2.4 – 75	3.7 – 47	2.7 – 252	5.1 – 263

BC: Black carbon, CO: Carbon monoxide, NO: Nitric oxide, NO₂: Nitrogen dioxide, NO_x: Nitrogen oxides, O₃: Ozone, T: Temperature, N: Number of hours with observations, SD: Standard deviation, IQR: Inter Quartile Range, Min: Minimum observation, Max: Maximum observation

Table 2.

Hourly averages of wind speed, temperature and traffic counts measured at the near-road continuous instrumentation. September 8, 2014 to January 5, 2015.

		Wind Speed (mph)	Temp (C)	Traffic Count*
	N	2343	2343	1920
Total	Mean	2.4	13	12300
9/8–1/5	SD	1.4	7	5780
	IQR	1.4 – 3.2	8 – 19	2180 – 19000
	Min-Max	0 – 7.0	−5 – 30	1400 – 20700
	N	226	226	288
September	Mean	2.3	22	12600
9/1–9/30	SD	1.0	3	5800
	IQR	1.6 – 3.0	19 – 23	7300 – 17300
	Min-Max	0 – 5.2	14 – 29	1560 – 20700
	N	744	744	504
October	Mean	2.0	19	12500
10/1–10/31	SD	1.3	6	5800
	IQR	0.9 – 3.0	15 – 23	7500 – 17100
	Min-Max	0 – 6.0	4 – 30	1500 – 20200
	N	676	676	456
November	Mean	2.6	9	12100
11/1–11/30	SD	1.6	6	5780
	IQR	1.4 – 3.6	5 – 14	6870 – 17080
	Min-Max	0 – 7.0	−5 – 24	1400 – 20000
	N	591	591	672
December	Mean	2.5	9	12100
12/1 – 12/31	SD	1.2	5	5780
	IQR	1.7 – 3.1	6 – 13	6680 – 17000
	Min-Max	0.04 – 6.9	−1 – 22	1490 – 20600

^*Traffic count data were collected at a location on I-75/I-85 1.5 miles south of the measurement location using Automatic Traffic Records

During the sampling period, CO and NO₂ hour maximums at both sites remained below the hourly national standards of 35 ppm for CO and 100 ppb for NO₂ despite the prominent wind direction being from the east. The maximum hourly concentration at the NR DRIVE and NR GIT sites were 1.9 ppm (CO) and 93.8 ppb (NO₂), and 2.2 ppm (CO) and 51.6 ppb (NO₂), respectively (Table 1). Due to high concentrations that skewed the distribution causing a non-normal distribution, sites were compared using a Spearman’s rank correlation. The NR DRIVE site on average measured lower CO and NO_x as well as higher NO₂ than those measured at the NR GIT location (Table 1). Temporal variability in CO and NO₂ hourly concentrations between the two sites lead to a Spearman’s correlation of 0.18 (CO) and 0.72 (NO₂). Both sites captured the morning and evening increase in TRAP concentrations in the same hour; however the average diurnal profiles show lower CO and NO₂ concentrations measured at the NR DRIVE site and the higher NO₂ concentrations measured at the NR GIT site (Fig. 2). Less dispersion at the NR GIT site due to vegetation drove different mixing rates with ozone, leading to differences between the NO and NO₂ concentrations at the NR DRIVE and NR GIT sites. This is also partly shown by the closer NO_x concentrations with the mean (standard deviation) concentration of 50 ppb (35 ppb) and 57 ppb (34 ppb) at the NR DRIVE and NR GIT sites, respectively, and the higher NO_x Spearman’s correlation between the two sites of 0.72. The lower NO₂ concentrations at the NR GIT site suggests less ozone titration occurs where reduced dispersion occurs.

Figure 2.

Diurnal profiles for CO, NO₂, NO, NO_x, BC, and O3 for the NR GIT and NR DRIVE sites from September 8, 2014 to January 5, 2015.

At the NR DRIVE and NR GIT sites, the mean (standard deviation) BC concentrations were 1.6 (1.3) ug m⁻³ and 1.7 (1.2) ug m⁻³. While the BC measurement at the NR GIT site was only operational during November and December, similar monthly averaged concentrations were observed at both sites during these months. Diurnally, the BC concentration at both sites followed a similar trend with a morning peak from 9am to 11am and an afternoon minimum at 5pm (Fig. 2). The BC diurnal trend mimicked the NO trend with a bimodal distribution observing a maximum concentration in the morning 1.5 times greater than the evening peak. These pollutant distributions differed from the more balanced bimodal distributions of the CO and NO₂ diurnal profiles, which observed similar peak concentrations in the morning and evening.

To understand the impact vehicle emissions have on local concentrations, the near-road measurements were also compared to the urban background and rural measurements around Atlanta, GA. Measurements for the urban background concentrations were collected at the highly instrumented, long-term, Jefferson St. site, part of the Southeastern Aerosol Research and Characterization (SEARCH) network (Blanchard et al. 2013a; Edgerton et al. 2005, 2006; Hansen et al. 2003; Hansen et al. 2006; Liu et al. 2005; Solomon et al. 2003) located 2.3km west of the near-road sites. Average (maximum) hourly concentrations for urban background CO and NO₂ were 266 ppb (1732 ppb) and 12.6 ppb (94.4 ppb). Also part of the SEARCH network, the Yorkville site located about 40 miles northwest of Atlanta provides rural background pollutant concentrations. The average (maximum) hourly concentrations for rural CO and NO₂ were 175 ppb (524 ppb) and 2.2 ppb (26.9 ppb) respectively. Based on the difference in the means for the measurements from September 8, 2014 to January 5, 2015, the regional background contributed about 28% to the NR GIT site CO measurements, and the urban emissions contributed 15% to the NR GIT site CO measurements. The highway vehicle emissions were a significant source of the CO concentration measured in the near-road environment contributing the remaining 57% to the measured CO concentration at the NR GIT site. In contrast, the regional background was a smaller percentage of the NO₂ concentration measured in the near-road environment (11%) and the urban background contributed to about 53% of the NR GIT NO₂ measurements. City scale regulations for NO₂ would help overall exposure since NO₂ highway emissions contribute only 35% to the near-road measurements.

Single pollutant concentrations measured in the near-road environment are commonly used as key indicators of the impact vehicle emissions have on local air quality within the near-road microenvironment. In order to assess the impacts from vehicle emissions in the near-road environment, CO and NO₂ were commonly measured to represent primary vehicle emissions (HEI 2010). The near-road measurements for both key TRAPs were elevated above the Atlanta urban and rural background concentrations suggesting traffic emissions contribute to elevated concentrations in the near-road environment. The NR DRIVE site concentrations for CO and NO₂ were 35% and 57% higher than the urban background concentrations. This is consistent with other EPA Near-road Monitoring Network sites in 2014, which reported mean NO₂ concentrations ranging from 9 to 24 ppb (DeWinter et al. 2018; EPA 2016). While traffic emissions contribute to the elevated levels of primary pollutants in the near-road environment, other urban and region sources contribute a significant fraction affecting their use as tracers for vehicle emissions. Overall, the pollutant levels measured at the DRIVE study location, an open site with good dispersion characteristics, showed a relatively small roadside increment and low impact of the 16-lane interstate highway compared to historic near-road field data, especially those measured from more enclosed sites (i.e. street canyon locations). Nevertheless, the pollutant levels we measured during this extensive monitoring period were consistent with measurements from the US EPA’s near-road monitoring network pollutant trends analysis and emissions estimates.

Assessment of traffic volume and meteorological factors impacting roadside concentrations

Traffic count and meteorological conditions are key factors driving near-road TRAP concentrations. Average weekday traffic data during the sampling period on the interstate demonstrated three peaks with the morning and evening rush hour events as well as a mid-day peak at about 2pm (Figure 3a). Instead of a common bimodal traffic count distribution (Baldauf et al. 2012; Batterman et al. 2015), this segment of highway had a consistently high traffic volume with vehicle counts rising quickly from 5am to 7am and more slowly until reaching a maximum at 3pm. Vehicle count slowly decreased from 3pm to 7pm and quickly dropped reaching a minimum at 3am. The daily trend was consistent across all four months of the study with variability in the travel behavior based on weekday (Figure 3b, Figure S3) and weekend (Figure 3c, Figure S3). Traffic volume on the weekend displayed no peak in the morning or evening, further highlighting the significance of both work commuting trips and the use of the highway for other daily trips.

Figure 3.

Hourly average traffic count of (a) total, (b) weekday, and (c) weekend variability from September 1, 2014 to December 31, 2014

The mean normalized diurnal profiles of the key TRAP species, vehicle counts, and meteorological conditions driving dispersion showed the extent of daily variation for each pollutant and factor for the NR DRIVE site and the NR GIT site (Figure 4). The normalized CO, NO₂, and BC concentrations had similar diurnal profiles with a morning concentration peak at 10am, an evening peak at 10pm, and minimum concentrations observed at 3am and 4pm. The normalized hourly O₃ concentration measured at the NR DRIVE site had a maximum concentration at 6pm and a minimum at 10am. Since the minimum primary pollutant concentrations occurred during the hours with maximum vehicle counts, this showed that highway traffic count alone is a poor indicator of diurnal pollutant levels observed at the DRIVE study location.

Figure 4.

Hourly mean (a) NR DRIVE concentration data, (b) NR GIT concentration data, and (c) traffic parameter data normalized by mean. (d) NR DRIVE concentration normalized by mean and multiplied by mixing height. (e) NR GIT concentration normalized by mean and multiplied by mixing height. Data from September 8, 2014 to January 5, 2015. TCNT: Traffic count, MH: Mixing height, SPD: Traffic highway speed.

The average vehicle speed and corresponding congestion patterns (Figure 4c) showed highway traffic speeds remaining high throughout the night and reaching a minimum at 6pm during the evening rush hour period. As traffic counts rose at 6am, corresponding reductions in mean traffic speeds were observed from about 70 to 40mph by 8am. While traffic counts remained high throughout the day, mean traffic speeds increased from 45mph at 10am before dropping to 15mph at 5pm. Since vehicle emissions rates remain fairly constant above 20mph (Barth et al.) and traffic counts remain elevated throughout the day, diurnal emissions trends would suggest the highest vehicle emissions occur between 2pm and 8pm when measured species concentrations are lowest.

Mixing height data was generated by the WRF model for the 4km grid including the two central near-road locations, and varied diurnally. The mixing height remained low (about 260m) until about 7am, increased during the day until reaching a maximum height at about 3pm (typically about 1000m), and decreased until approximately 10pm. Mixing height as well as ozone formation driven by photochemical activity resulted in a peak for both between 1pm and 6pm. As the mixing height increased in the morning, TRAP concentrations decreased reaching a minimum concentration at 3pm while the traffic count was reaching a maximum. By multiplying the concentration by the mixing height (Figure 4d and 4e), the increase during mid-day when the mixing height was greatest showed the emissions increased with traffic count, but mixing height drove daily dynamics leading to minimum concentrations mid-day.

The two near-road sites observed trends in primary TRAPs explained by rapid pollutant dispersion, particularly associated with the increase in convective mixing and increased wind speeds. Even at near-road sites, the diurnal convective mixing and wind speed had the dominant impact of TRAP concentrations (Fig. 4). The high impact of meteorological factors compared to traffic count suggests a change in fate and transport properties affecting near-road concentration variability. This change influences the applicability of traffic count as a mobile source tracer in quantifying exposure to traffic emissions. Our results were consistent with recent findings from other near road study (Hilker et al. 2019; Sofowote et al. 2018; Wang et al. 2018). Further, an implication of the changing near-road environment is that future exposure studies aimed at characterizing health impacts of mobile emissions will need to consider different approaches for determining the mobile source contribution to ambient concentrations of single pollutants.

The pollutant diurnal patterns at the two near-road network sites were consistent with prior studies showing elevated concentrations of primary traffic pollutants occurring during morning rush hours when the atmospheric mixing is weak and emissions are high, then concentrations decreasing at the boundary layer increases (Grosjean 1983; HEI 2010; Menut et al. 2012). During the evening rush hours, the concentrations again increased and remained high throughout the night in spite of the greatly reduced emissions due to the low boundary layer. Both sites observed the typical bi-modal diurnal profile for the primary traffic-related air pollutants, but did not observe the same diurnal profile for vehicle count, which is often used as a proxy of exposure to traffic pollutants in health effect studies (Batterman et al. 2015). Traffic counts on I-75/I-85 through metro Atlanta exhibited distinctive patterns of rising sharply in the early morning consistent with the beginning of the morning rush hour and reached a consistent peak vehicle count of approximately 20,000 vehicles per hour from 10am to 4pm. The differences in the diurnal profile patterns between the vehicle counts and primary traffic pollutant concentrations highlights the predominant role meteorology and its influence on vertical dispersion have on the impact of traffic hotspots on adjacent areas.

Assessment of combined traffic volume and meteorological factors leading to roadside concentrations

As part of the DRIVE study, exposures were modeled by integrating hourly concentrations. Further, the NAAQS for both NO₂ and CO have one-hour components. Thus, there is a need to link local and regional factors to hourly concentrations. This can be done using more complex dispersion modeling approaches or statistical methods. Statistical modeling can help identify if different factors influence the observations at the three near-road locations differently. Here, linear mixed modeling evaluates associations between pollutant concentrations and multiple possible contributing predictors to assess factors that drive the temporal variability observed at the different near-road sites. The regression coefficients for the models developed for the NR DRIVE site and the NR GIT site were compared to assess whether site differences along the same road segment can lead to significant differences in pollutant dynamics (Table 3). Significance of a factor was determined by a p-value less than 0.05.

Table 3.

Regression coefficients from multivariate models examining associations between multiple factors and hourly pollutant concentrations at the NR DRIVE and NR GIT sites from September 8, 2014 to January 5, 2015.

	NR DRIVE BC		NR GIT BC		NR DRIVE CO		NR GIT CO		NR DRIVE O3
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Late Evening (9pm-12am)	0.23*	(0.04, 0.42)	0.54*	(0.29, 0.79)	6.08	(−27.25, 39.41)	47.63*	(6.97, 88.29)	1.31*	(0.24, 2.38)
Early Morning (1–5am)	0.24*	(0.03, 0.44)	0.52*	(0.24, 0.79)	−8.52	(−44.92, 27.88)	−22.45	(−66.23, 21.34)	2.05*	(0.89, 3.20)
Morning Rush Hour (6–9am)	0.39*	(0.13, 0.64)	0.70*	(0.37, 1.03)	−11.98	(−55.88, 31.92)	−1.59	(−54.11, 50.94)	1.85*	(0.42, 3.27)
Mid Day (10am-3pm)	0.14	(−0.04, 0.32)	0.49*	(0.25, 0.72)	−27.31	(−58.23, 3.61)	25.87	(−11.83, 63.57)	1.54*	(0.56, 2.53)
Temperature (C)	0.01*	(0.00, 0.02)	−0.02*	(−0.04, −0.00)	2.35*	(0.74, 3.97)	−4.00*	(−6.13, −1.86)	0.40*	(0.35, 0.45)
Relative Humidity (%)	0.00	(0.00, 0.01)	0.00	(−0.01, 0.01)	−0.37	(−1.31, 0.58)	−0.33	(−1.59, 0.93)	−0.22*	(−0.26, −0.19)
Wind Speed (mph)	−0.20*	(−0.26, −0.14)	−0.19*	(−0.27, −0.12)	−40.75*	(−50.09, −31.40)	−45.35*	(−57.35, −33.34)	2.54*	(2.24, 2.84)
Northerly Wind	−0.27*	(−0.47, −0.08)	−0.74*	(−1.28, −0.20)	−36.78*	(−70.63, −2.92)	−90.25*	(−134.28, −46.23)	1.83*	(0.73, 2.93)
Easterly Wind	−0.73*	(−0.95, −0.51)	−1.25*	(−1.79, −0.70)	−59.43*	(−97.25, −21.61)	−191.16*	(−240.46, −141.87)	5.06*	(3.83, 6.29)
Southerly Wind	−0.30	(−0.76, 0.17)	-	-	−7.17	(−83.98, 69.64)	−222.118	(−324.75, −119.47)	3.44*	(0.83, 6.04)
Weekend	−0.61*	(−0.97, −0.26)	−0.81*	(−1.40, −0.22)	−11.84	(−74.45, 50.77)	−9.13	(−92.74, 74.49)	2.34*	(0.40, 4.28)
Traffic Count (per 1,000)	0.06*	(0.04, 0.07)	0.00*	(0.06, 0.09)	8.60*	(6.56, 10.64)	0.02*	(19.11, 24.09)	−0.31*	(−0.37, −0.24)
Mixing Heights (per 100 meters)	−0.03*	(−0.05, −0.01)	0.00	(−0.04, 0.02)	−3.41	(−7.12, 0.30)	−6.39*	(−11.14, −1.64)	0.47*	(0.06, 0.34)

	NR DRIVE NO		NR GIT NO		NR DRIVE NO2		NR GIT NO2		NR DIRVE NOx		NR GIT NOx
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Late Evening (9pm-12am)	4.73*	(1.89, 8.65)	11.13*	(6.97, 15.28)	0.27	(−1.63, 2.17)	0.34	(−0.75, 1.43)	5.00*	(0.35, 9.65)	11.39*	(6.59, 16.19)
Early Morning (1–5am)	5.42*	(3.69, 12.46)	11.92*	(7.35, 16.48)	−2.22*	(−4.26, −0.17)	−1.00	(−2.20, 0.19)	3.18	(−1.82, 8.18)	10.85*	(5.58, 16.13)
Morning Rush Hour (6–9am)	7.53*	(−0.30, 5.97)	14.63*	(9.22, 20.03)	−1.85	(−4.30, 0.60)	−0.35	(−1.76, 1.05)	5.65	(−0.34, 11.64)	14.18*	(7.94, 20.41)
Mid Day (10am-3pm)	3.25*	(2.13, 9.47)	7.81*	(3.97, 11.65)	−1.95*	(−3.71, −0.20)	0.18	(−0.82, 1.19)	1.28	(−3.00, 5.57)	7.94*	(3.50, 12.37)
Temperature (C)	−0.10	(−0.28, 0.04)	−0.83*	(−1.02, −0.63)	0.04	(−0.05, 0.13)	−0.05*	(−0.11, −0.00)	−0.06	(−0.28, 0.16)	−0.89*	(−1.11, −0.66)
Relative Humidity (%)	0.21*	(0.13, 0.32)	0.12*	(0.01, 0.24)	−0.15*	(−0.20, −0.09)	−0.13*	(−0.16, −0.19)	0.06	(−0.07, 0.20)	0.00	(−0.14, 0.13)
Wind Speed (mph)	−1.83*	(−2.93, −1.02)	−2.97*	(−4.15, −1.79)	−5.06*	(−5.60, −4.53)	−2.39*	(−2.70, −2.07)	−6.88*	(−8.19, −5.56)	−5.33*	(−6.70, −3.96)
Northerly Wind	−5.64*	(−29.36, −12.68)	−9.16*	(−13.56, −4.76)	−4.39*	(−6.36, −2.41)	−3.33*	(−4.50, −2.16)	−10.04*	(−14.88, −5.20)	−12.45*	(−17.54, −7.36)
Easterly Wind	−19.00*	(−23.16, −15.18)	−19.72*	(−24.62, −14.82)	−10.21*	(−12.41, −8.00)	−5.89*	(−7.19, −4.59)	−29.25*	(−34.65, −23.85)	−25.53*	(−31.20, −19.85)
Southerly Wind	−20.66*	(−9.29, −2.14)	−26.59*	(−36.83, −16.35)	−3.23*	(−7.84, 1.39)	−3.01*	(−5.72, −0.30)	−23.95*	(−35.24, −12.65)	−29.55*	(−41.41, −17.70)
Weekend	−9.06*	(−14.80, −3.15)	−13.09*	(−20.32, −5.86)	−4.97*	(−8.13, −1.82)	−2.44*	(−4.44, −0.45)	−14.03*	(−21.95, −6.10)	−15.40*	(−24.01, −6.80)
Traffic Count (per 1,000)	1.09*	(0.88, 1.30)	0.00*	(2.10, 2.60)	0.50*	(0.39, 0.61)	0.00*	(0.36, 0.50)	1.58*	(1.30, 1.86)	2.76*	(2.47, 3.05)
Mixing Heights (per 100 meters)	−0.35	(−0.75, 0.05)	0.00	(−0.80, 0.16)	−0.64*	(−0.86, −0.42)	0.00*	(−0.45, −0.20)	−0.99*	(−1.53, −0.45)	−0.64*	(−1.19, −0.08)

^*All covariates were included simultaneously in the model from each pollutant of interest. Est. Estimate of Coefficient; 95% CI-95% Confidence Interval; Unit for BC: μg m⁻³; Unit for CO, NO, NO₂, NO_x, and O₃: ppb; *Significant with a P-value of 0.05

During the 2014 study period, the NR DRIVE site and the NR GIT site regression coefficients for BC and NO concentrations were positively associated with late evening to morning rush hour period (9pm to 9am) when the traffic count and mixing height were low and beginning to increase (Table 3). Also for both sites, wind direction and increasing wind speed were significantly associated with decreases in all the primary pollutants (BC, CO, NO, NO₂, and NO_x), indicative of dispersion away from the pollutant source (Table 3 and Figure S4). Additionally, weekend days showed an association with a significant decreasing concentration for all pollutants (NO, NO₂, and BC) except CO. While many of the factors were significant for both sites, temperature was negatively significant for NO, NO₂, and NO_x only at the NR GIT site. Temperature was also significantly negative at the NR GIT site for BC and CO, while positively significant for the NR DRIVE site. The NR DRIVE site is located in an open parking lot indicative of increased photochemical reactions compared to the NR GIT site, which was located within a tree barrier along the highway. The regression models also highlight the diminishing predictive power of traffic count on near-road pollutant levels. While the coefficient for traffic count was significant at both near-road sites and for all pollutants, the magnitude of the coefficient was low and therefore was not a key factor driving temporal variability of pollutant concentration at the study domain. Nevertheless, while traffic counts alone has been found not to be an important predictor of TRAP concentrations, a recent near road study completed in Canada has identified heavy-duty diesel traffic volume as an important predictor (Sofowote et al. 2018; Wang et al. 2018).

Changing Near-road Environment

Given the relatively short duration that we analyzed for the current study, caution should be taken regarding the generalizability of the findings for making long-term inferences regarding how the near-road environment has changed over the last decade. A state operated, routine monitoring site (SDK), while not associated with the EPA Near-Road Monitoring Network, provides a long-term concentration data set located within 650m of the I-285 bypass around metro Atlanta with an AADT of about 140,000. The hourly average (maximum) concentrations from September 8, 2014 to January 5, 2015 for CO and NO₂ were 323 ppb (1259 ppb) and 9.7 ppb (57.5 ppb), respectively. From 2000 to 2010, the observed CO and NO₂ concentration at the SDK site decreased 33% (496 to 345 ppb) and 42% (18.1 to 13.7 ppb), respectively. For the same period from 2000 to 2010, the CO concentration decrease was 50% (561 to 282 ppb) and 9% (180 to 163 ppb) at the urban (JST) and rural (YRK) background sites. Similarly, the NO₂ concentration decrease was 30% (21.9 to 15.3 ppb) and 60% (5.4 to 2.2 ppb), respectively. Decreased mobile emissions have contributed to a decrease in near-road concentrations for primary traffic-related air pollutants; however, a decrease in overall concentrations locally and regionally have also contributed to the decrease in the near-road environment. Further, the difference in the rate of decrease from sources contributes to the changing near-road environment and how to characterize near-road exposure.

As observed here and elsewhere, near-road TRAP concentrations are less elevated above background levels than prior near-road studies around the country (Beckerman et al. 2008; Sarnat et al. 2008), suggesting that future studies will need to consider different approaches for characterizing the spatial gradients and exposures to mobile sources. Initial results from the EPA Near-road Monitoring Network support these results. Of the 61 near-road sites active in 2015, only five hourly concentrations exceeded the hourly NO₂ standard of 100 ppb and at all the sites the 98^th percentile of the daily 1-hour maximum was below the standard (DeWinter et al. 2018). The NO₂ concentration average across the country ranged from about 9 ppb to 30 ppb.

Nationally, estimates of on-road mobile source emissions of NO_x have decreased about 50% since 2004 and emissions of CO show that 2014 levels are about 49% of those in 2004 and 25% of those in 1994 (Dallmann and Harley 2010). In the metro Atlanta area, long term analysis data from the urban background site part of the SEARCH network shows that CO, NO_x, and BC levels have decreased by 350 ppb (7.2% per year), 35 ppb (7.3% per year), and 1.25 μg m⁻³(6.8% per year), respectively, from 1999 to 2011. Source apportionment analysis at this site indicated mobile source-related PM_2.5 decreased by about half during the same period (Blanchard et al. 2013b). These declining trends are expected to continue in the future as fleet turnover to newer vehicles continues, new mobile source emissions controls are introduced and additional policy interventions are implemented.

Conclusion

Elevated traffic-related air pollutants are linked to adverse health effects and often epidemiological health analysis rely on only a few monitoring locations to quantify exposure of individuals in the near-road environment. Our comparison of two near-road monitors along the same highway segment showed site-to-site differences influencing pollutant concentrations. Site characteristics can contribute to localized concentrations and this affect was observed when comparing the two near-road sites in this study. The two sites were 325m apart and both within a meter of a heavily trafficked highway. While the two sites observed similar diurnal and temporal variability, site differences including increased vegetation around one location reduced dispersion in comparison to the location by an open, asphalt parking lot. Reduced dispersion lead to higher CO and NO concentrations and lower NO₂ concentrations over the study period. These site differences contributed to different pollutant dynamics along the same highway segment highlighting the importance of site location. The near-road pollutant concentrations measured show a reduced impact from the highway sources with levels less elevated above background concentrations than in prior studies. The regression models also highlighted the diminishing predictive power of traffic count on near-road pollutant levels. These decreased concentrations indicate the effectiveness of mobile source emission controls leading to a decreased relative contribution from vehicles to urban air pollution. This finding indicates a changing near-road environment that will affect future approaches to characterizing vehicle emission hotspots and their impacts on exposures.

Highlights.

• Lesser impact from highway source on pollutant levels compared to prior studies

• Low predictive power of traffic count on near-road pollutant levels

• Results indicative of the effectiveness of mobile source emission controls

• Use of observations from a single near-road site can lead to biases in assessing exposures to mobile emissions