Measurements of Enhanced Turbulent Mixing near Highways

1618

JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY

VOLUME 51

Measurements of Enhanced Turbulent Mixing near Highways* MARK GORDON, RALF M. STAEBLER, JOHN LIGGIO, PAUL MAKAR, SHAO-MENG LI, JEREMY WENTZELL, GANG LU, PATRICK LEE, AND JEFFREY R. BROOK Science and Technology Branch, Environment Canada, Toronto, Ontario, Canada (Manuscript received 30 August 2011, in final form 2 April 2012) ABSTRACT In August and September of 2010, measurements of turbulent fluxes and turbulent kinetic energy were made on highways in the Toronto area (Ontario, Canada). In situ turbulence measurements were made with a mobile laboratory while driving on the highway with traffic. Results demonstrate that the turbulent kinetic energy (TKE) spectrum is significantly enhanced on and near the highway by traffic for frequencies above 0.015 Hz. The decay of TKE with distance behind vehicles is well approximated by power-law curves. The strongest increase in TKE is seen while following heavy-duty trucks, primarily for frequencies above 0.7 Hz. From these results, a parameterization of on-road TKE enhancement is developed that is based on vehicle type and traffic-flow rate. TKE with distance downwind of the highway also decays following a power law. The enhancement of roadside TKE is shown to be strongly dependent on traffic flow. The effect of vehicle-induced turbulence on vertical mixing was studied by comparing parameterized TKE enhancement with the typical TKE predictions from the Global Environmental Multiscale weather forecast to predict the potential increase in vertical diffusion that results from highway traffic. It is demonstrated that this increase in TKE by traffic may be locally significant, especially in the early morning.

1. Introduction Traffic emissions have a substantial effect on air quality on an urban and regional scale. Turbulence affects how these emissions mix with the surrounding air, which determines the three-dimensional distribution of the traffic pollutants downwind of roadways. This is especially relevant to human health given the proximity of urban populations to major roadways. For example, 45% of the population in Toronto, Ontario, Canada, live within 500 m of an expressway or within 100 m of a major road (Health Effects Institute 2010) and approximately 16% of American households are within 100 m of a highway having four or more lanes (U.S. Census Bureau 2008). In a study of pollution in southern Ontario, Makar et al. (2010) found that the model known as A Unified

* Supplemental information related to this paper is available at the Journals Online website: http://dx.doi.org/10.1175/JAMC-D11-0190.s1.

Corresponding author address: Ralf Staebler, Science and Technology Branch, Environment Canada, 4905 Dufferin St., Toronto, ON, M3H 5T4. E-mail: [email protected] DOI: 10.1175/JAMC-D-11-0190.1

Regional Air-quality Modeling System (AURAMS; Moran et al. 1998) tended to overpredict primary (i.e., emitted) pollutant concentrations near the ground in grid squares with highways, and speculated that the cause may be the lack of vehicle-induced turbulence in the driving meteorological conditions. Parameterizations of near-road turbulence enhancement exist (e.g., Kalthoff et al. 2005), but there is a need for parameterization of traffic-induced turbulence enhancement that is based on in situ on-road measurements. There are three mechanisms that increase mixing near roadways (Rao et al. 1979). First, flow modification due to elevation of the roadway and structures such as dams and dividers can enhance turbulent mixing. Second, the asphalted surface of the road can enhance heat flux relative to surrounding areas, especially in agricultural settings. These two effects are respectively termed [following Wang and Zhang (2009)] structural road-induced turbulence (S-RIT) and thermal road-induced turbulence (T-RIT). Third, the motion of traffic on the roadway causes vehicle-induced turbulence (VIT). VIT has been studied using models (Danard 1972; Baümer et al. 2005; Wang and Zhang 2009; Wang et al. 2011), wind tunnels (Eskridge and Thompson 1982; Kastener-Klein et al. 2000), and field experiments (Rao et al. 1979; Chock 1980; Kalthoff et al.

SEPTEMBER 2012

GORDON ET AL.

2005). Rao et al. (1979) measured eddy sizes of several meters resulting from VIT, and Chock (1980) demonstrated the influence of moving vehicles on spectra, variances, covariances, and turbulence intensity. In a field study on an autobahn in Germany, Kalthoff et al. (2005) measured a 50%–70% increase in TKE downwind of the motorway during perpendicular winds (primarily for TKE frequencies of greater than 0.1 Hz). On-highway measurements of VIT are rare, with chasing studies (measurements made while following moving vehicles) typically focusing on gas and particle concentration measurements (Kittelson et al. 2004; Pirjola et al. 2004; Wehner et al. 2009; Carpentieri and Kumar 2011). Wake turbulence behind vehicles has been studied in wind tunnels (Eskridge and Thompson 1982) and using computational fluid dynamics (CFD) modeling (Baker 2001). The only study to the authors’ knowledge in which turbulence was measured during chasing was conducted in 1999 with results presented in Rao et al. (2002) and Hosker et al. (2003). In these studies, a trailer equipped with sonic anemometers was towed by a full-sized van on an airport runway. A significant increase in TKE was seen, especially at high frequencies. TKE and the wake deficit were found to decrease with distance from the vehicle. Rao et al. suggest that additional field measurements in vehicle wakes are required, with sufficient data to form ensemble averages for different vehicle types and conditions. The study presented here attempts to develop parameterizations of RIT and VIT as a function of traffic flow and composition that can be added to a meteorological model equipped with the appropriate input fields. In situ turbulence measurements were made on a highway with a mobile laboratory: the Canadian Regional and Urban Investigation System for Environmental Research (CRUISER). This vehicle-chasing study is the first in which the effects of vehicle type and following distance on turbulence and mixing are investigated. In addition to the in situ measurements on the highway, the variation of turbulence perpendicular to the highway was also studied at a near-road location. In the following sections the mobile laboratory, its instrumentation, and the instrumentation at the near-road location are described, spectra of turbulence are compared for multiple locations, turbulence in the wakes of vehicles is compared for different vehicle types and downstream distances, and a parameterization of on-road VIT is developed. The production of TKE near the highway is compared with traffic-flow observations in an attempt to estimate the relative contributions of S-RIT, T-RIT, and VIT and to parameterize the total near-road TKE on the basis of traffic flow and time of day. These results are then used to investigate the potential enhancement of

1619

mixing by traffic through comparison with weatherforecast-model TKE levels and implied vertical diffusion enhancements.

2. Methods The Fast Evolution of Vehicle Emissions from Roadways (FEVER) subproject, which is part of the Advancing Local-scale Modeling through Inclusion of Transportation Emission Experiments (ALMITEE) project, took place between 16 August and 17 September 2010. The objective was to measure traffic emissions and turbulence on and near highways in the area surrounding Toronto. Measurements were made using both fixed instrumentation at a roadside location and the mobile laboratory, either on the highway with traffic or on a side road perpendicular to the highway. This manuscript investigates the enhancement of turbulence and mixing as a result of traffic and individual vehicles. For details of traffic emissions, pollutant mixing, and the broader scope of the FEVER project, the reader is referred to Gordon et al. (2012) and Liggio et al. (2012).

a. On-road measurements To measure in situ enhanced turbulence that results from individual vehicles, the mobile laboratory was driven on Highways 400 and 401 near Toronto. Both of these highways are multilane, separated-direction roadways. Driving took place on Highway 400 between Toronto and Port Severn (approximately 125 km north of Toronto) and on Highway 401 between Toronto and Darlington (100 km east of Toronto). The mobile laboratory was driven on six days between 20 August and 15 September 2010, resulting in a total of approximately 14 h of measurements. The mobile laboratory was outfitted with two 3D sonic anemometers (model CSAT3 from Campbell Scientific, Inc., of Logan, Utah) and an aircraft turbulence probe (model AIMMS-20 from Aventech Research, Inc., of Barrie, Ontario, Canada). The sonic anemometers and the turbulence probe were mounted to a frame at a height of 3.0 m, as shown in Fig. 1. The anemometers sampled 3D wind speed (u, y, and w) and temperature T at a rate of 20 Hz for the full duration of the study. The AIMMS20 aircraft probe was not used for turbulence measurements owing to the difficulty in calibrating the instrument for the modified pressure field around the vehicle. The instrument’s high-frequency global positioning satellite (GPS) sensors and inertial motion sensing were utilized to measure the mobile laboratory position, its speed UV at a rate of 5 Hz, and the three components of acceleration (ax, ay, az) at a rate of 40 Hz. Another accelerometer (Summit Instruments, Inc., of Akron, Ohio)

1620


VOLUME 51

FIG. 1. Photograph of CRUISER, the mobile laboratory, demonstrating the mounting locations of the two sonic anemometers and the aircraft turbulence probe.

was mounted on the right side of the frame (near the right-side anemometer) and recorded the three components of acceleration at a rate of 10 Hz. During the design phase of the anemometer support structure, the flow distortion caused by the mobile laboratory while driving (including the mounting frame) was studied using CFD simulations (Unicell Body Company, Inc., of Buffalo, New York). For a vehicle speed of 100 km h21, the modeled upward flow at the location of the anemometers caused by flow distortion around the vehicle was approximately 2 m s21 and there was no significant increase in modeled TKE at that location. Crosswinds of up to 8.3 m s21 (30 km h21) had a negligible effect on the modeled updraft velocity and TKE enhancement. To determine the following distance and vehicle type during highway driving (also known as ‘‘chasing’’), a digital video camera was installed on the dashboard of the vehicle that recorded video at 1 frame per second. The video postprocessing and calibration techniques used to determine the following distance xF at a rate of 1 Hz are outlined in the supplemental online material. Chased vehicles were classified as a passenger car, medium-sized truck, or heavy-duty (HD) truck (either 10 or 18 wheeled) by manually viewing the video footage and noting the changes in followed vehicle types.

b. Near-road measurements A side road perpendicular to Highway 400 (43.9948N, 79.5838W) was chosen for roadside measurements between 17 August and 17 September 2010. The side road is shown in Fig. 2. This location was chosen for reasons of high traffic volume, with an average of 110 000 vehicles per day (MTO 2009), a lack of surrounding structures and pollution sources, and a nearly north–south alignment, which is perpendicular to the predominantly westerly winds. The highway at this location is six lanes wide (25 m across from the lane edges), with a 1-m-high barrier at the center, and drainage channels (;2 m deep) on either side. Traffic at this location is primarily due to commuters heading south into Toronto in the morning and returning north in the evening; on Fridays and weekends, the road is heavily traveled by those heading to cottage areas north of Toronto. Vegetation in the surrounding area is predominantly agricultural, with some trees lining the side roads. Winds at this location were predominantly from the west, although south and southeasterly winds occurred during the study. A 3-m roadside tower (Fig. 2) was installed 9.5 m east of the highway edge (22 m from the highway center), which supported a CSAT3 sonic anemometer at a height of 3.0 m relative to the road surface, a radiometer (model

SEPTEMBER 2012

1621

GORDON ET AL.

FIG. 2. The roadside study site with measurement locations. Perpendicular distances from Highway 400 are given in parentheses. The locations of measurements are shown, comprising wind speed components (u, y, and w), TKE e, heat flux H, shortwave radiation S, traffic flow F, wind speed U, and wind direction u. The mobile laboratory was either parked at the location shown or was driven along Side Road 18.

LI-200 from Li-Cor, Inc., of Lincoln, Nebraska) at a height of 2.3 m, and a temperature/relative humidity sensor (Campbell Scientific model HMP35C) at a height of 2.0 m. The sonic anemometer on the tower sampled at a rate of 10 Hz. The tower instruments were used to calculate TKE e, heat flux H, and incoming shortwave radiation S for the duration of the study. An airquality monitoring system (airpointer from Recordum Messtechnik GmbH of Mo¨dling, Austria) was used to measure wind speed U and direction u, averaged every minute, 34 m east of the highway center. A traffic camera (Miovision Technologies, Inc. of Kitchener, Ontario, Canada) was mounted at the roadside overlooking the highway. The recordings were postprocessed by the manufacturer to measure traffic-flow rate (vehicles per minute F) with a manufacturer-specified accuracy of greater than 95%. The traffic flow was classified as passenger cars Fc medium-sized vehicles Fm or HD vehicles Ft. These classifications were further subdivided as either northbound FN or southbound FS. The mobile laboratory was used at this roadside location to drive transects along the east and west lengths of the side road, perpendicular to the highway. This was typically done for a few hours at a time, spanning peak traffic during the morning and/or the evening. More than 44 h of transect driving were done at this site on 17 days,

resulting in 8 morning and 14 evening sessions. The transects are described in further detail in Gordon et al. (2012).

c. TKE measurement TKE measurements were calculated as e 5 0:5(u92 1 y92 1 w92 ) ,

(1)

where u92 , y92 , and w92 are the velocity fluctuations computed over the time period T 5 fsN, with recording frequency fs and number of values per period N. The variances in Eq. (1) are often written with an overbar to denote averaging; it is omitted herein for simplicity, and all variances and wind speeds (e.g., u92 or u) presented are averages over time period T. For TKE, variance, and flux calculations, all wind velocity components were rotated following the method of Wilczak et al. (2001) to align u with the mean wind direction (giving y 5 w 5 0). The time period must be long enough to capture the lowfrequency variation in the TKE but short enough not to capture unrelated variation such as the diurnal cycle of TKE. In atmospheric measurements a period of 30 min is typical. During vehicle-chasing measurements, there are frequent changes in the surrounding environment, including

1622


VOLUME 51

FIG. 3. TKE with varying averaging period recorded by the mobile laboratory during highway driving (black) and at the roadside tower downwind of the highway (NW winds; red line) and upwind of the highway (SW winds; blue line) winds. The TKE is normalized by the value for the longest period.

the following distance xF (the distance between the mobile laboratory and the vehicle being followed by the mobile laboratory). To capture a representation of turbulence due to a single vehicle type at a relatively constant following distance, it was necessary to use a relatively short averaging time of T 5 10 s, which does not capture the larger time scales of the turbulence. To determine the underestimation of TKE that is due to this short averaging time, measurements were used during a portion of chasing while the mobile laboratory followed an HD vehicle for 13 min (with following distance varying between 12 and 35 m). TKE was calculated by splitting the time series into a number of blocks with period T and then averaging the TKE of all of the blocks for that value of T. The resulting values of TKE (normalized by the TKE calculated with T 5 13 min) are shown in Fig. 3. At T 5 10 s, the averaged TKE is 68% of the TKE calculated with T 5 13 min. This is compared with the variation of TKE calculated from measurements at the roadside tower during a northwest (NW) wind at 1000 eastern daylight time (EDT) 14 September and at 1000 EDT 21 August during a southeast (SE) wind. During an NW wind, the tower is downwind of the highway, whereas during an SE wind it is upwind of the highway (Fig. 2). At this location, a period of T 5 10 s would only give 51% of the total TKE during NW winds and 54% during SE winds. The large difference between the normalized TKE at T 5 10 s for stationary and chasing measurements is likely due to the smaller (higher frequency) eddies generated by the HD vehicles, which contribute

a larger portion of the total TKE on the highway. Hence, we estimate that our measurements capture approximately 2/ 3 of the TKE during highway driving, because of the necessary short sampling time. Measurements of TKE calculated using T 5 10 s are identified as 10-s TKE e10s to differentiate from TKE e calculated with T 5 30 min.

d. Measurement uncertainty It is possible that vehicle vibration during driving could lead to false anemometer measurements, as the measurements are relative to the reference frame of the mobile laboratory, which will vibrate because of uneven road conditions. Accelerometers mounted in the aircraft probe recorded the three components of acceleration (ax, ay, az) at a rate of 40 Hz. Because horizontal vehicle motion was recorded by GPS (at 5 Hz), we are only concerned here with acceleration az caused by vertical vibration. The acceleration az was integrated to give vertical velocity wz,a due to vehicle motion. For a sample of highway driving while following 20 m behind an HD vehicle (moving at 27 m s21), the standard deviation of vertical velocity measured by the anemometer was 2.09 m s21 while the standard deviation of wz,a was sw 5 0.25 m s21, suggesting a contribution of approximately 12% from the vehicle motion. At slower speeds (during roadside measurements), the standard deviation of wz,a was sw 5 0.12 m s21, approximately one-half of the highway value. Although the contribution of vibration to the variation of vertical velocity may be significant, the contribution of vibration to the total TKE measurement is much less:

SEPTEMBER 2012

1623

GORDON ET AL.

TABLE 1. Description of data used for comparison of power spectra shown in Fig. 4 (with corresponding numbered scenarios). Upwind and downwind are relative to the highway (Fig. 2), T is the length of time over which the spectra are calculated, U is the mean wind speed (with vehicle motion subtracted), e is the average TKE, and w92 is the variance of the vertical velocity. No.

Measurement platform

Location

1 2 3 4 5 6 7

Tower Tower Mobile laboratory parked Mobile laboratory parked Mobile laboratory transect Mobile laboratory Mobile laboratory

Upwind Downwind 22 m Downwind 300 m Downwind 300 m Downwind 50–550 m On highway behind car On highway behind truck

Date (2010) Time (EDT) T (min) U (m s21) e (m2 s22) w92 (m2 s22) 21 Aug 14 Sep 31 Aug 18 Aug 1 Sep 14 Sep 14 Sep

0.03 m2 s22 during highway chasing and 0.007 m2 s22 during roadside transects. These are negligible values relative to the observed TKE.

3. Results a. Turbulence spectra To investigate the relative scales of turbulence captured by the measurements, seven scenarios were chosen

0800 0800 1200 0000 1748 1247 1257

30 30 30 30 9 1 1

4.13 2.53 1.49 0.74 2.18 4.14 3.94

1.21 2.65 0.70 0.23 1.88 4.48 14.53

0.34 0.66 0.23 0.04 0.31 1.32 5.53

to study the turbulence produced under various conditions on and near the highway. The details of each scenario are listed in Table 1. Short sampling times were necessary during highway driving (scenarios 6 and 7) to ensure a nearly uniform vehicle speed (UV ’ 26 m s21) and following distance (between 15 and 25 m). The power spectra of the vertical velocity variance w92 and TKE e are compared for the seven scenarios in Fig. 4. Frequencies fj are normalized with the measurement

FIG. 4. Power spectral densities of (top) w92 and (bottom) total TKE for various scenarios (Table 1). Label UW indicates upwind of the highway, and DW is downwind of the highway. Spectra are normalized such that the total of all of the points in the shared range (0.01 , fn , 5) for each curve gives the total variance. The 22/ 3 power law is shown as a dashed line.

1624


height za (53.0 m) and the wind speeds U shown in Table 1. In the case of highway driving, frequencies were normalized by U 1 UV. Normalized spectra S* are presented that relate to the actual spectra S as n

B

j50

j5A

å fj S 5 å fj S* 5 w92 ,

(2)

where A and B are the shared frequency range such that fA 5 0.005 and fB 5 1. This normalization allows comparison of the magnitude of the spectra, which are derived from different averaging periods and measurement frequencies. Peak normalized frequencies are in the range from 0.05 to 0.9 for w92 while TKE spectra peaks are at generally lower frequencies, between 0.02 and 0.3, with the exception of the turbulence spectra produced while following a truck during highway driving, which does not peak within the range shown. As expected, higher TKE is associated with higher wind speeds, with weaker TKE during the night when wind speeds were lower (scenario 4), and higher TKE at the side road during upwind measurements (scenario 1) and during the transect (scenario 5), when wind speeds were higher. On 14 September (scenario 2) although the wind speed recorded at the tower downwind of the highway is lower than the wind speed on 21 August (scenario 1) when the tower is upwind of the highway, the TKE and w92 are higher, demonstrating enhancement of turbulence by VIT or RIT. Kalthoff et al. (2005) present TKE power spectra (as individual u, y, and w components) for stationary towers upwind and downwind of a German motorway (with wind perpendicular to the highway). The TKE power spectra measured by their downwind tower (3.0 m from the highway edge) are identical to the TKE power spectra measured by their upwind towers for frequencies below 0.05 Hz and are consistently higher for f of greater than 0.05 Hz (by approximately one-half of a decade on the logarithmic scale). A similar pattern is seen in our tower measurements (Fig. 4) for the upwind and downwind conditions. The TKE power spectra measured when the tower is downwind of the highway are nearly identical to the TKE power spectra measured by the towers when they are upwind of the highway below fn ’ 0.01 ( f ’ 0.015 Hz) and are increasingly higher above that frequency (also reaching approximately one-half of a decade on the log scale). The w92 downwind and upwind spectra measured at the tower diverge at a higher frequency ( fn ’ 0.2), suggesting that the horizontal component of eddies caused by VIT contributes more to the TKE than the vertical component of eddies in the midrange frequencies. This is consistent with the aspect ratios of vehicles, which would result in longer length scales in

VOLUME 51

the horizontal dimensions. The higher divergence frequency for the TKE spectra in the Kalthoff et al. study may be due to the different structures of the highways, because the four-lane motorway in the Kalthoff et al. study was raised by 1 m, whereas the six-lane highway in this study was not raised significantly. This obstruction would result in a vertical compression of the flow and an acceleration of the wind over the highway, possibly leading to higher-frequency turbulence. The most striking difference between the turbulent spectra shown in Fig. 4 is during highway driving. While following a car, there is a significant increase in turbulence relative to the roadside measurements in the range 0.01 , fn , 0.3. While following a truck, the increase for fn . 0.1 ( f ’ 0.7 Hz) is much greater and the decay at higher frequencies does not follow the 22/ 3 power law expected in the inertial subrange. This is likely due to the creation of eddies in this frequency range (fn . 0.1) that balance the typical dissipation that is seen under normal atmospheric conditions. Hence, these measurements likely underestimate the actual TKE production. Because the upper limit of this frequency range is greater than the highest measured frequency (20 Hz), the amount of this underestimation is unknown. A similar increase in the TKE power spectra due to VIT is seen in the results of Rao et al. (2002). In Rao et al. (2002), three sonic anemometers were mounted at a height of 1.83 m on a trailer behind a 2.1-m-high fullsized van. The distance between the van and the anemometers ranged from 0.6 to 4.9 m. An additional three anemometers were mounted on the trailer at a height of 3.7 m to measure ambient conditions outside of the van wake. The TKE power spectra of the lower anemometers (in the vehicle wake) were higher than the those of the ambient conditions across the measured spectral range (0.08 , f , 5 Hz), with a greater difference at higher frequencies. At frequencies above f ’ 1 Hz, the spectra were nearly flat or were increasing slightly with f. This result is very similar to the nearly flat spectra for fn . 0.1 (f . 0.7 Hz) observed while following a truck in this study (Fig. 4).

b. Vehicle wakes on the highway The 10-s averages of TKE e10s during highway driving were binned by vehicle type and following distance xF. Results for each vehicle type (car, midsize, and HD vehicles) represent multiple vehicles that were followed one at a time. The results demonstrate a power-law decrease in TKE with following distance, especially for HD vehicles. When the TKE values are further subdivided by vehicle and wind speed combined (UM 5 UV 1 U), the TKE at lower speeds is lower than the TKE at higher speeds for equivalent following distances. Here, we present

SEPTEMBER 2012

1625

GORDON ET AL.

FIG. 5. Quartiles (for each vehicle type from bottom to top: 25th percentile, median, and 75th percentile) of binned normalized 10-s TKE with the normalized following distance behind cars, midsized vehicles, or HD vehicles. The speed includes both the mobile laboratory speed and the wind speed parallel to the vehicle motion. Also shown are least squares fits (colored thick dashed lines) and parameterizations based on scaled wind-tunnel measurements (Eskridge and Thompson 1982; solid black line) and trailer-based measurements (Rao et al. 2002; black dashed line). The Rao et al. (2002) parameterization is extrapolated from following distances of less than 2h. 2 10-s TKE data normalized by UM and following distance normalized by an approximated average vehicle height h (1.4, 2, and 4.1 m for cars, midsize vehicles, and HD vehicles, respectively). Because there are uncertainties associated with turbulence measurements at low wind speeds, data with UM of less than 5 m s21 were not used in the analysis. Normalized TKE is compared with xF/h in Fig. 5. Least squares fits are shown for each vehicle type, with fit statistics listed in Table 2. Standard errors of the fit parameters give the 68% confidence intervals. The median values demonstrate better agreement with power-law fits in this range than with linear or exponential decays (not shown), which is consistent with the analytical results of Eskridge and Hunt (1979) and the measurements of Rao et al. (2002). The TKE generated by trucks demonstrates a strong dependence on normalized following distance. The TKE

data measured behind cars and midsize vehicles both decrease with increasing following distance, but there is much less decrease in TKE with distance when compared with that observed when following HD vehicles. The decay of TKE with distance stops beyond 24h for the HD vehicles and 60h for the midsize vehicles (both near 100 m). This could imply that the vehicle wakes have no influence beyond this distance, although in both cases the normalized TKE remains higher than that of cars at the equivalent following distance. This may be due to increased uncertainty in distance measurements in that range (see the online supplemental material), which would obscure the dependence of TKE on distance; the same effect is not seen in the results while following cars, however. Eskridge and Hunt (1979) developed a model of the TKE in the wake of a vehicle that is based on a self-preserving

2 2 TABLE 2. Least squares fit to ln(e10s /UM ) 5 a 1 b ln(xF /h) [or e10s /UM 5 exp(a)(xF /h)b] of normalized 10-s TKE as a function of normalized following distance. The standard errors (here SE) in a and b and the number n of 1-Hz data are listed. The fits of Eskridge and Thompson (1982) and Rao et al. (2002) are listed for comparison.

Study

Vehicle type

exp(a)

b

r2

SE(a)

SE(b)

n

This study This study This study Eskridge and Thompson (1982) Rao et al. (2002)

Cars Midsize Heavy duty Midsize Midsize

0.0092 0.0125 0.0821 0.0051 0.0099

20.23 20.34 20.92 21.20 20.19

0.052 0.120 0.642

0.037 0.025 0.014

0.0109 0.0083 0.0065

8306 11923 11227

1626


VOLUME 51

FIG. 6. Average 10-s TKE with following time behind cars, midsized vehicles, or HD vehicles. The dashed line shows estimated background levels.

solution. According to the model, the TKE decays with following distance as (x/h)23/2. Eskridge and Thompson (1982) measured the wake behind a block-shaped vehicle (modeled after a midsized car) in a wind tunnel. They modified the model of Eskridge and Hunt on the basis of results from a 1/ 32-scaled model. Scaling their results to full size and assuming an approximate drag 2 5 0.0051(xF /h)21.2. coefficient of CD 5 0.5 gives e/UM This is shown in Fig. 5 for comparison. Changing the 0:5 . drag coefficient changes the factor of 0.0051 as CD Although the block model was based on a midsized vehicle, the scaled, nondimensional TKE from the Eskridge and Hunt experiments is nearly equal to the scaled, nondimensional TKE for following HD vehicles from this study at a following distance near 3h. The predicted decay of TKE with distance is steeper than the parameterization of this study that is based on HD vehicles and is below most measurements of this study (for all vehicles) for x . 25h. The lower TKE values at higher following distances may be due to interference among multiple vehicles in the on-road study, which would not be a factor in the windtunnel study with a single vehicle. In the Rao et al. (2002) study discussed in section 3a, TKE was measured at trailing distances of 0.6, 2.7, and 4.9 m, for vehicle speeds between 4.4 and 22 m s21. A parameterization was developed by determining a logarithmic best fit of normalized TKE to the normalized fol2 5 0.0099(xF /h)20.19. The lowing distance, resulting in e/UM Rao et al. (2002) fit is shown in Fig. 5. Despite the much shorter trailing distances of the Rao et al. (2002) study, the normalized TKE is within the range of normalized TKE for the car and midsize vehicles in this study.

c. On-road turbulence enhancement To estimate the total contribution of vehicles to onroad turbulence, the TKE is binned by the time behind

the followed vehicle, which is calculated as tF 5 xF/UM. The resulting averages of 10-s TKE in raw units are shown for cars, midsize vehicles, and HD vehicles in Fig. 6. Without normalization, the maximum values of TKE occur near 0.75 s for cars and HD vehicles and the TKE is lower closer to the vehicles (lower tF). This is likely due to a bias of lower speeds during very close following distances. Using these temporal distributions of TKE behind vehicles, the average TKE enhancement of F vehicles per unit time can be calculated for each vehicle type as De 5 F

ð‘ 0

(e 2 ebg ) dt ,

(3)

where ebg is the background on-road TKE, which is estimated as 2.4 m2 s22 on the basis of the averages values for tF . 8 s (Fig. 6). Integrating the average binned values of TKE shown in Fig. 6 gives an average on-road TKE enhancement of De 5 Ic Fc 1 Im Fm 1 It Ft ,

(4)

where subscripts c, m, and t correspond to cars, midsize vehicles, and HD vehicles, respectively, and I is the value of the integral in Eq. (3), which is determined numerically from the average binned values to give Ic 5 2.7 m2 s21, Im 5 0.7 m2 s21, and It 5 18 m2 s21. Because of the variability of the measurements, the standard errors of these values are high, estimated as 1.4, 1.1, and 2.1 for Ic, Im, and It, respectively. Although the value of I for midsize vehicles is smaller than the value of I for cars, the values are not significantly different within a 68% confidence interval. As an example calculation of on-road TKE enhancement, the peak traffic flow at the near-road measurement site (discussed in section 2b) occurred between 1700 and 1800 EDT and was approximately

SEPTEMBER 2012

GORDON ET AL.

1627

FIG. 7. (a) Average TKE and traffic flow (sum of northbound and southbound) and (b) sensible heat flux and incoming solar radiation, binned by hour. The TKE, heat flux, and solar radiation are separated according to wind direction to differentiate between measurements downwind of the highway (subscript D), and measurements upwind of the highway (subscript U). (c) Differences in hourly downwind and upwind average solar radiation and heat flux.

F 5 2.2 s21, which is composed of 89.9% cars, 4.8% midsize vehicles, and 5.3% HD vehicles. Substituting these values into Eq. (4) gives an on-road TKE enhancement of 3.4 m2 s22.

d. Roadside turbulence enhancement Although simultaneous upwind and downwind roadside measurements of TKE were not available during this study, the enhancement of TKE can be inferred by comparing measurements made during westerly winds (when the instruments were downwind of the highway) with measurements made during easterly winds (when the instruments were upwind of the highway). Westerly and easterly are defined here as being perpendicular to the highway 6608. Values of incoming solar radiation S as well as heat flux H and TKE e were binned by hour of day and separated by wind direction u to give the downwind and upwind diurnal patterns shown in Figs. 7a–c. For all of the analyses in this section, TKE and heat flux were determined with T 5 30 min to include lower-frequency contributions. During

the night (2100–0500 EDT), the TKE eD downwind of the highway was higher than the TKE eU upwind of the highway. Because the turbulence created by solar heating of the road (T-RIT) is not a factor at night, this result was likely due to S-RIT and/or VIT. During the day, there was a significant increase in the heat flux downwind of the highway relative to upwind of the highway (Fig. 7b), with more than double the heat flux resulting from flow across the road. This may be due in part to a bias in the weather conditions (clear vs cloudy) during downwind measurements relative to upwind measurements. The difference between the average downwind and upwind diurnal solar radiation (SD 2 SU) is compared with the difference between the average downwind and upwind heat flux (HD 2 HU) in Fig. 7c. Conditions were generally cloudier during upwind measurements, resulting in generally positive differences. There is moderate correlation between SD 2 SU and HD 2 HU (correlation coefficient squared r2 5 0.52); the difference in heat flux is generally larger than the difference in solar radiation,

1628


VOLUME 51

however, suggesting that some of the variation is likely due to surface heating of the road. Although it is difficult to determine a direct relationship between heat flux and TKE, Kaimal et al. (1976) relate the heat flux to variance of vertical wind speed as zg 2/3 w9T9 . w92 (z) 5 1:8 T

(5)

Between 1400 and 1500 EDT (when the road is hottest), using median values of temperature and heat flux, Eq. (5) gives w92 (at 3 m) 5 0.19 m 2 s22 downwind of the highway and 0.07 m2 s22 upwind of the highway. This is as compared with measured median values for the same time period of 0.27 m2 s22 downwind of the highway and 0.05 m2 s22 upwind of the highway. The modeled w92 from Eq. (5) is 0.12 m2 s22 higher for winds crossing the highway, whereas the measured w92 is 0.22 m2 s22 higher for winds crossing the highway. Hence, Eq. (5) predicts that 53% (0.12 m2 s22/0.22 m2 s22) of the total measured vertical mixing for this time period is due to T-RIT, with the remaining 47% being due to S-RIT and VIT. Although the upwind TKE eU increases during the day, the effect is more pronounced in the diurnal pattern of the downwind TKE eD. The diurnal variation of traffic flow F at the roadside location is shown in Fig. 7a. The increase in downwind TKE also starts earlier and ends later in the day, generally following the pattern of traffic flow, as opposed to solar radiation. The traffic flow shown in Fig. 7a includes all vehicle types. The median values of northbound HD-vehicle flow (not shown) increase linearly from a minimum of 1 min21 at 0300 EDT to a maximum of 2 min21 at 1100 EDT and then decrease linearly back to the minimum at 0300 EDT. The difference between the median hourly values of TKE (eD – eU) is compared with median hourly total traffic flow in Fig. 8. There is a strong correlation between the TKE difference and the total traffic flow for both northbound and southbound lanes (r2 5 0.78). The correlation is stronger (r2 5 0.88), however, if only the northbound-lane traffic flows are used (i.e., traffic nearest the roadside-tower anemometer). A least squares fit of De 5 eD 2 eU 5 a 1 bFN

(6)

gives a 5 0.14 m2 s22 and b 5 0.70 m2 s21. The intercept of 0.14 m2 s22 provides an estimate of the turbulence created by the road structure and the 1-m-high dividing wall at the road center (S-RIT). Kalthoff et al. (2005) measured TKE simultaneously upwind and downwind of a motorway and determined a logarithmic least squares fit of the TKE difference to

FIG. 8. The difference between average diurnal TKE values downwind of the highway (westerly winds) and upwind of the highway (easterly winds) as a function of the traffic flow (all vehicle types) for northbound traffic flow (open circles) and northbound plus southbound traffic flow (plus signs). A least squares fit to the northbound traffic flow is shown (red line). Equation (7) is shown with two values of wind speed U (blue lines).

the wind speed. This least squares fit was then used to normalize the TKE difference to give two linear least squares fits to traffic flow for passenger cars or HD (‘‘commercial’’) vehicles. We combine these two parameterizations, weighted by car and truck flow fractions, to give eD 2 eU UV2

!,

U UV

1:2 5

Fc Fc 1 Ft

1

(ac 1 bc Fc )

Ft Fc 1 Ft

(at 1 bt Ft ) ,

(7)

where the subscripts c and t denote the fit parameters and traffic flow for cars or HD vehicles, respectively. The fraction of HD traffic during the Kalthoff et al. study was 6.2% as compared with 4.0% during this study. The TKE differences calculated from Eq. (7) are shown in Fig. 8 for wind speeds of 1 and 3 m s21 and vehicle velocities of 28 m s21 (120 km h21). This range of wind speed roughly corresponds to 1 standard deviation from the average wind speed during the study. The data including both sides of the road are within the range of the Kalthoff et al. parameterization, demonstrating relatively good agreement between the two studies. The value of S-RIT during the Kalthoff et al. study can be calculated from Eq. (7) (with U 5 1 m s21 and Fc 5 Ft 5 0) as eD 2 eU 5 0.06 m2 s22, which is less than one-half of the value found in this study. This difference could be due to the lack of a solid barrier on the motorway, which is instead bounded by guardrails.

SEPTEMBER 2012

1629

GORDON ET AL.

FIG. 9. The 10-s TKE measured at the roadside tower binned by distance from the road following the wind direction, showing averages (plus signs), medians and quartiles (boxes), and 10th and 90th percentiles (whiskers). Average values from the mobile laboratory during transects are also shown (green line). A least squares fit to the median TKE measurements for xw/hT , 24 is shown (blue line).

e. Evolution of turbulence with distance from the highway The wind speed measurements from the roadside tower (22 m from the highway center) can provide an indication of how turbulence changes with horizontal distance from the highway. Although the tower is stationary, it is possible to investigate changes in a Lagrangian sense by estimating the distance a parcel of air travels along the wind direction from the highway to the tower location. For consistency with the highway measurements described above and because of the variability in wind direction, e10s is calculated from the tower measurements with T 5 10 s. The distance xw from the road along the wind trajectory is taken from the highway centerline. For dimensional consistency, TKE is normalized by wind speed and distance is normalized by the anemometer height, hT 5 3.0 m. The normalized 10-s TKE binned by xw/hT is shown in Fig. 9. As with the vehicle-chasing measurements, very low wind speeds (U , 0.1 m s21) were not used in the analysis. As with Fig. 5, there is a power-law decrease in TKE with distance until a background level is reached near xw 5 24hT (72 m). Average values of e/U2 during times when the tower is upwind of the highway (easterly winds) are slightly lower at 0.39 (median of 0.35) as compared with 0.44 (median of 0.38) at xw /hT 5 32. As with the vehiclechasing measurements, the independence of e/U2 with xw /hT in this range may be due to increased uncertainty in the calculation of distance from the highway when the wind direction is at a higher angle relative to the

highway. For xw , 24hT only, a least squares fit to the median values gives e/U 2 5 1:94(xw /hT )20:53 ,

(8)

with a correlation squared of r2 5 0.99. The tower measurements are compared with the mobile laboratory 10-s TKE measurements made during transects on the side road perpendicular to the highway. For the mobile laboratory measurements, the distance along the wind trajectory xw was determined using the airpointer (34 m from the highway center) U and u. The average normalized TKE values measured on the mobile laboratory demonstrate good agreement with the average normalized TKE values measured at the roadside tower, although there is much more noise in the data because of the smaller sample size. The TKE averages during transects are slightly higher than the tower measurements, most likely as a result of a bias in sampling only during morning and afternoon rush hours. The mobile laboratory TKE measurements also continue to decrease with normalized distance, suggesting background levels of turbulence are not reached until distances near 50hT (150 m).

4. Discussion Enhancement of TKE by the highway and traffic may have a significant effect on the vertical diffusion of trafficemitted pollutants. To investigate this enhancement, we

1630


VOLUME 51

FIG. 10. Diffusion coefficient [Eq. (9)] at a 3-m height with GEM-modeled TKE and with TKE enhanced by traffic flow according to Eq. (4) at the Highway 400 stationary site. The percent increase in K due to the traffic-enhanced TKE is also shown.

used output of the Global Environmental Multiscale (GEM) model (Coˆte´ et al. 1998; Mailhot et al. 2006), which was run for the southern Ontario region as part of the Border Air-Quality Study and Meteorology Study (BAQS-Met), as described in Makar et al. (2010). The model was run for the time period between 17 June and 10 July 2007 with 2.5-km grid spacing, nested from a driving 15-km grid-spacing simulation. The lowest vertical level of the GEM meteorological model is 42 m, giving a model volume for the lowest level of 0.26 km3. Enhancement of TKE by the highway and traffic may not significantly increase the average TKE in such a large volume; vertical diffusion could be enhanced above the road itself, however. This in turn could lead to enhanced dilution of traffic emissions into the next vertical level. The higher wind speeds at higher levels could also lead to increased horizontal transport of the pollutants. Vertical diffusion of pollutants is governed in the GEM model (Mailhot et al. 1998) by K 5 cle1/2 /u ,

(9)

where c is a constant, l 5 k(z 1 z0) is a mixing length for the statically neutral case with surface roughness length z0, and u(Ri) is a stability function determined by the Richardson number Ri. To determine the potential impact of VIT on local stability and mixing, the GEM-modeled TKE values e and stability functions u of the lowest level at the location of the near-road Highway 400 site were output as hourly averages for the entire model run (17 June– 10 July 2007). Although the model estimates are not time coincident with the current study, the use of time averages allows an order-of-magnitude enhancement to be

investigated, from the original model levels of turbulence to those resulting from both meteorological influences and the presence of vehicles (VIT). The diffusion coefficient calculated with the hourly modeled values of e and u is shown in Fig. 10. The parameterization of on-road TKE enhancement (VIT) in Eq. (4) is a function of the vehicle-flow rates of cars, midsize vehicles, and HD vehicles, whereas the parameterization of near-road TKE enhancement [Eq. (6)] is a function of total vehicle flow only. Using the measured composition of traffic at peak flow (1700–1800 EDT: 89.9% cars, 4.8% midsize, 5.3% HD), Eq. (4) gives De/F 5 3.4 m2 s21. This can be compared with the slope b 5 0.7 m2 s21 of Eq. (6), which is the roadside enhancement of TKE due to VIT only (neglecting S-RIT). The value of near-road TKE enhancement per vehicle is lower by a factor of 4.9 than the on-road value. Although the lower value is quantitatively consistent with the horizontal distribution of TKE discussed in section 3e (Fig. 9), a qualitative comparison is difficult because the normalized TKE follows a power-law decay with distance along the wind trajectory while the roadside enhancement of TKE from Eq. (6) is due to all winds within 608 of perpendicular from the highway. For winds within 608 of perpendicular from the highway, the distance between the road center and the near-road tower along the wind trajectory can vary from 22 to 44 m. Using the power-law fit of Eq. (8), the value of normalized TKE at the center of the northbound lanes (xw 5 6.3 m) is e/U2 5 1.3. At the roadside tower location along a perpendicular trajectory (xw 5 22 m), e/U2 5 0.67. At the roadside tower location along a trajectory 608 to perpendicular (xw 5 44 m), e/U2 5 0.47. If a background level of e/U2 5 0.38 is assumed, the ratio of on-road to roadside TKE enhancement is

SEPTEMBER 2012

GORDON ET AL.

between 3.2 and 10.2, which is in agreement with the ratio of on-road to roadside TKE enhancement of 4.9 derived from Eqs. (4) and (6) above. Although there is substantial variability in the measurements and range of expected values and uncertainty in the choice of background level, this demonstrates a relative agreement among the three measurement techniques and analysis discussed in sections 3c (on-road enhancement), 3d (roadside enhancement), and 3e (roadside horizontal distribution). The values of e, De, and u were used to calculate the diurnal vertical diffusion coefficients at the site location. The diffusion coefficients were calculated with z 5 3 m, using both the GEM-average e values and the GEMaverage e values modified by Eq. (4), with the diurnal median traffic flows. Because we are concerned here with mixing directly above traffic, the traffic flow is calculated for each lane. Northbound and southbound traffic flows are divided into three lanes each, and the average traffic flow per lane is used in Eq. (4). The enhanced diffusion coefficient is compared with the unmodified diffusion coefficient in Fig. 10. The increase in the diffusion constant due to the inclusion of traffic-induced TKE varies between 10% and 25% with the exception of the earlymorning hours, when it increases to more than 80%. This early-morning increase may be crucial to modeled ozone production, since this corresponds to the time of active chemistry, when the production of ozone is initialized by the radical reaction chemistry due to nitrous acid photolysis.

5. Conclusions These results demonstrate that TKE is significantly enhanced on and near the highway by traffic. The increased TKE is due to contributions in the higherfrequency ranges (f . 0.015 Hz). This supports the results of Kalthoff et al. (2005), in which increases in TKE were reported for f . 0.05 Hz. The strongest increase in TKE over background levels was observed while following HD trucks, for frequencies above 0.7 Hz. As was reported in the results of Rao et al. (2002) when following a van at a fixed distance, the spectrum is relatively flat above 1.3 Hz and no inertial subrange is observed. This suggests that the production of turbulence that results from the truck motion balances the dissipation in that high-frequency range. The decay of TKE with distance behind vehicles is well approximated by a power-law curve that is based on vehicle type, with exponents of 20.92 for trucks, 20.34 for midsize vehicles, and 20.23 for cars. Although this is the first study to measure TKE with following distance in a realistic setting, the results for midsize vehicles are

1631

in good agreement with the measurements of Rao et al. (2002), which were made in a more controlled environment. From these measurements, a typical traffic-flow rate of 100 cars per minute results in an on-road TKE enhancement of 4.5 m2 s22 while a typical HD vehicle flow rate of five trucks per minute results in an on-road TKE enhancement of 1.5 m2 s22. These TKE measurements were made with 10-s periods and, hence, exclude TKE contributions for less than 0.1 Hz. It is estimated that the contribution of frequencies of less than 0.1 Hz would account for approximately 1/ 3 of the TKE. In following HD vehicles, measurements suggest that there is also a significant but unknown contribution to the total TKE at frequencies above 20 Hz. It is not possible to estimate the value of this contribution with this dataset. Using 30-min TKE measurements (which account for frequencies , 0.1 Hz), the enhancement of TKE at the side of the highway shows a strong correlation with traffic flow. On the basis of a parameterization of near-road TKE enhancement with traffic flow, the average S-RIT for this study is estimated as 0.14 m2 s22. The parameterization predicts that a typical peak daily northbound flow of 100 vehicles per minute will result in a TKE increase due to VIT of 1.2 m2 s22. Although it is difficult to distinguish between TKE enhancement due to T-RIT and that due to VIT, the higher values of heat flux measured for winds passing over the road suggest that T-RIT may contribute more than one-half of the vertical mixing during the warmest part of the day. The TKE decayed with downwind distance from the highway following a power-law curve with an exponent of 20.53. The increase in TKE due to the highway traffic extended as far as 150 m downwind. Although it is difficult to compare the on-road and near-road measurements because of the necessarily different analyses at each location, the three measurement techniques of onroad TKE enhancement, near-road TKE enhancement, and horizontal distribution of TKE downwind of the highway are shown to be in good agreement. Adding the parameterized generation of VIT to the GEM weather forecast model average meteorologicalinduced TKE demonstrates the potential for a significant increase in vertical mixing due to traffic-generated turbulence. At the study location, the increase in the vertical diffusion coefficient due to VIT can be near 80% in the early morning and between 10% and 25% throughout the day. This increased vertical diffusion would result in an increased fraction of pollutants being mixed into the next vertical model layer, which could explain the tendency of the AURAMS to overpredict pollutant concentrations near the ground in grid squares with highways (Makar et al. 2010). Potential improvements to the

1632


model could include partitioning pollutant emissions into highway-based and non-highway-based components, with a parameterized TKE increase as a function of traffic flow for the highway-based emissions. Acknowledgments. This work was supported by the Natural Science and Engineering Research Council of Canada and was funded by the Science and Technology Branch of Environment Canada and the Particles and Related Emission Project (PERD) C11.008, which is a program administered by Natural Resources Canada. REFERENCES Baker, C. J., 2001: Flow dispersion in ground vehicle wakes. J. Fluids Structures, 15, 1031–1060. Baümer, D., B. Vogel, and F. Fiedler, 2005: A new parameterisation of motorway-induced turbulence and its application in a numerical model. Atmos. Environ., 39, 5750–5759. Carpentieri, M., and P. Kumar, 2011: Ground-fixed and on-board measurements of nanoparticles in the wake of a moving vehicle. Atmos. Environ., 45, 5837–5852. Chock, D. P., 1980: General Motors sulfate dispersion experiment. Bound.-Layer Meteor., 18, 431–451. Coˆte´, J., S. Gravel, A. Me´thot, A. Patoine, M. Roch, and A. Staniforth, 1998: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373– 1395. Danard, M. B., 1972: Numerical modeling of carbon monoxide concentrations near highways. J. Appl. Meteor., 11, 947– 957. Eskridge, R. E., and J. C. R. Hunt, 1979: Highway modeling. Part I: Prediction of velocity and turbulence fields in the wake of vehicles. J. Appl. Meteor., 18, 387–400. ——, and R. S. Thompson, 1982: Experimental and theoretical study of the wake of a block-shaped vehicle in a shear-free boundary flow. Atmos. Environ., 16, 2821–2836. Gordon, M., R. M. Staebler, J. Liggio, S.-M. Li, J. Wentzell, G. Lu, P. Lee, and J. R. Brook, 2012: Measured and modeled variation in pollutant concentration near roadways. Atmos. Environ., 57, 138–145. Health Effects Institute, 2010: Traffic-related air pollution. HEI Special Rep. 17, 386 pp. [Available online at http://pubs. healtheffects.org/getfile.php?u5553.] Hosker, R. P., Jr., K. S. Rao, R. L. Gunter, C. J. Nappo, T. P. Meyers, K. R. Birdwell, and J. R. White, 2003: Issues affecting dispersion near highways: Light winds, intra-urban dispersion, vehicle wakes, and the ROADWAY-2 dispersion model. NOAA Tech. Memo. OAR ARL-247, 197 pp. [Available online at http:// www.arl.noaa.gov/documents/reports/arl-247.pdf.] Kaimal, J. C., J. C. Wyngaard, D. A. Haugen, O. R. Cote´, Y. Izumi, S. J. Caughey, and C. J. Readings, 1976: Turbulence structure in the convective boundary layer. J. Atmos. Sci., 33, 2152–2169.

VOLUME 51

Kalthoff, N., D. Baümer, U. Corsmeiser, M. Kohler, and B. Vogel, 2005: Vehicle-induced turbulence near a motorway. Atmos. Environ., 39, 5737–5749. Kastener-Klein, P., R. Berkowicz, and E. J. Plate, 2000: Modelling of vehicle induced turbulence in air pollution studies for streets. Int. J. Environ. Pollut., 14, 496–507. Kittelson, K. B., W. F. Watts, and J. P. Johnson, 2004: Nanoparticle emissions on Minnesota highways. Atmos. Environ., 38, 9–19. Liggio, J., and Coauthors, 2012: Are emissions of black carbon from gasoline vehicles underestimated? Insights from near and on-road measurements. Environ. Sci. Technol., 46, 4819– 4828. Mailhot, J., and Coauthors, 1998: Scientific description of RPN Physics Library—Version 3.6. Recherche en Pre´vision Nume´rique Scientific Description, 197 pp. [Available online at http://collaboration.cmc.ec.gc.ca/science/rpn/physics/physic98. pdf.] ——, and Coauthors, 2006: The 15-km version of the Canadian Regional Forecast System. Atmos.–Ocean, 44, 133–149. Makar, P. A., and Coauthors, 2010: Mass tracking for chemical analysis: The causes of ozone formation in southern Ontario during BAQS-Met 2007. Atmos. Chem. Phys., 10, 11 151– 11 173. Moran, M. D., Dastoor, A. P., Gong, S.-L., Gong, W., and Makar, P. A., 1998: Conceptual design for the AES Unified Regional Air Quality Modelling System (AURAMS). Atmospheric Environment Service Air Quality Research Branch Internal Rep., 100 pp. MTO, 2009: Provincial highways: Traffic volumes—King’s highways, secondary highways, tertiary roads: 1988/2007. Ministry of Transportation Highway Program Planning Office Rep., 860 pp. Pirjola, L., and Coauthors, 2004: ‘‘Sniffer’’—A novel tool for chasing vehicles and measuring traffic pollutants. Atmos. Environ., 38, 3625–3635. Rao, K. S., R. L. Gunter, J. R. White, and R. P. Hosker, 2002: First field measurements of turbulence in the wake of a vehicle. Atmos. Environ., 36, 4337–4346. Rao, S. T., L. Sedefian, and U. H. Czapski, 1979: Characteristics of turbulence and dispersion of pollutants near major highways. J. Appl. Meteor., 18, 283–293. U.S. Census Bureau, 2008: American Housing Survey for the United States: 2007. U.S. Census Bureau Current Housing Reports Series H150/07. Wang, Y. J., and K. M. Zhang, 2009: Modeling near-road air quality using a computational fluid dynamics model, CFD-VIT-RIT. Environ. Sci. Technol., 43, 7778–7783. ——, A. DenBleyker, E. McDonal-Buller, D. Allen, and K. M. Zhang, 2011: Modeling the chemical evolution of nitrogen oxides near roadways. Atmos. Environ., 45, 43–52. Wehner, B., U. Uhrner, S. von Lo¨wis, M. Zallinger, and A. Wiedensohler, 2009: Aerosol number size distributions within the exhaust plume of a diesel and a gasoline passenger car under on-road conditions and determination of emission factors. Atmos. Environ., 43, 1235–1245. Wilczak, J. M., S. P. Oncley, and S. A. Stage, 2001: Sonic anemometer tilt correction algorithms. Bound.-Layer. Meteor., 99, 127–150.