As shown in Fig. 1, the FTIR and scintillometer instruments were set up on the south side of Highway 401 at 125 Resources Road (43.711∘ N, 79.543∘ W) in Toronto, Ontario, Canada. Our study was from 16 to 31 July 2015. The FTIR measurements were taken with a commercial open-path FTIR spectrometer (Open Path Air Monitoring System (OPS), Bruker, Germany). The infrared source is an air-cooled globar. The emitted radiation is directed through the interferometer where it is modulated, travels along the measurement path across the highway, reaches a retroreflector array that reflects the radiation back, travels back across the highway, and enters a Stirling-cooled mercury cadmium telluride detector (bistatic configuration). The FTIR spectrometer was set up on the roof of a building, about 8 m above the ground, while the retroreflector array was mounted on a mast at 4 m above ground level north of Highway 401. The distance between the spectrometer and retroreflector array was 310 m, resulting in a path length of 623.7 m, which includes 3.7 m of internal reflections. The length of the path that was directly over the highway was 117 m (Fig. 1).

In this study, spectra were measured at a resolution of 0.5 cm-1 with 250 scans co-added to increase signal-to-noise ratio, resulting in roughly a 1-minute temporal resolution. Before 23 July, 100 scan co-additions were used. At the beginning of the measurement period, a stray light spectrum was recorded by pointing the spectrometer away from the retroreflector. This stray light spectrum accounts for radiation back to the detector from reflections by internal parts inside the spectrometer, i.e., not from the retroreflector array, and was subtracted from all the measurement spectra before performing further analysis. Stray light affected retrieved final mixing ratios by < 3 % in this study. Spectral ranges for retrieval analysis in the Bruker software, OPUS_RS, for each target gas were chosen based on prominent absorption features of the target gas and spectral windows as found in previous studies as shown in Table 1. Reference spectra were fitted to the measured spectra using nonlinear curve fitting methods within the chosen window.

For each gas of interest, a reference file was made including spectra of target and interference gases. High-resolution reference spectra at 296 K and 1013.25 hPa were taken from the HITRAN database when available. For species not available in the HITRAN database, the reference spectra were taken from the PNNL database. Spectral ranges for fitting, interference gases, and detection limit based on Bruker's results for each pollutant retrieved in this work are listed in Table 1. Examples of measured spectra, model fit spectra in the optimum spectral ranges, and residuals are shown in Fig. S1 in the Supplement. Raw estimates of mixing ratios of gases of interest were retrieved assuming an ambient temperature of 296 K and air pressure of 1013.25 hPa. These values were then corrected for the actual temperature and pressure measured at the NAPS station using the ideal gas law.

The air temperature also has a secondary effect on the signature of the IR spectrum of individual gases. The population of the higher vibrational and rotational states increases as temperature increases. However, the sensitivity of temperature on those signatures depends on the individual gas and the range of temperature change. HITRAN and PNNL reference spectra at different temperatures are available for a limited number of species at 278, 298, and 323 K (Rothman et al., 1998, 2013; Sharpe et al., 2004; Johnson et al., 2010). Temperature-dependent reference files can be made in the OPUS_RS software to combine reference spectra at these three temperatures. To test the effect of temperature on the retrieved mixing ratio, spectra during the last 8 days of July were analyzed for NH3, CH4, CO, and CO2 by using these temperature-dependent reference files. The maximum difference in retrieved mixing ratio for the 45 ∘C range is 8.9 % for NH3, 4.2 % for CH4, 8.3 % for CO, and 4.1 % for CO2. Based on this test, we estimate that using reference spectra at standard temperature and pressure contributed to an uncertainty of less than 10 % in the final mixing-ratio results. Besides fitting errors and the effect of ambient temperature on reference spectrum, other environmental conditions may also contribute to uncertainties, such as interference from ambient water vapour.

Simultaneous long-path turbulence measurements were made using a boundary layer scintillometer (BLS 900, Scintec, Germany). The scintillometer receiver was set up next to the FTIR spectrometer, on the south side of Highway 401 (Fig. 1). The transmitter, with two disks of 924 LEDs emitting at 880 nm, was set up on the north side of Highway 401 just above the FTIR retroreflector. The mean height of the scintillometer path was 8 m above ground level. In this study, the LEDs were operated in continuous mode. Radiation is directed onto the photodiodes in the receiver, which quantify the turbulence-induced fluctuations in the optical refractive index between the transmitter and receiver. The theory of scintillometer measurements and calculations are included in the Supplement Sect. S2. Sensible heat flux (H), friction velocity (u∗), and Obukhov length (L) were calculated from scintillometer measurements at a 1 min resolution in this study.

z/L (where z is the height above the surface) is a surface-layer scaling parameter describing the dynamic stability of the surface layer (Stull, 2003). Negative z/L values indicate an unstable surface layer, while positive z/L values indicate a stable surface layer. The closer the value of z/L is to zero, the closer conditions are to neutral stability. In this work, H and z/L were used to determine the strength of turbulence and mixing in the surface layer (Fig. 2). Solar radiation data were taken from a York University weather station (http://www.yorku.ca/pat/weatherStation/index.php) situated about 9 km northeast of our site. We used the downwelling short wavelength radiation data to quantify cloudiness during the study. During the 16-day measurement period, only 17 July had some rain, and all other days were mostly sunny.

The NAPS program aims to provide accurate and long-term air quality data with uniform standards across Canada by coordinating the data collection from existing air quality monitoring networks (Galarneau et al., 2016). The first NAPS measurements were conducted in 1972, focusing on SO2 and particulate matter. Currently, SO2, CO, NO2, O3, and PM2.5 are continuously measured at more than 200 sites across Canada (Environment and Climate Change Canada, http://www.ec.gc.ca/rnspa-naps/default.asp?lang=En&n=8BA86647-1, last accessed 25 March 2017). The data shown in this study come from the NAPS trailer located right beside the FTIR path on the south edge of the Highway 401 (see Fig. 1). Pollutants monitored by this NAPS trailer include CO, NO, NO2, O3, SO2, and PM2.5 at 1-minute temporal resolution. The CO analyzer (model 48i TLE Enhanced Trace Level CO Analyzer, Thermo Fisher Scientific, USA) operates based on infrared absorption and gas filter correlation, the NOx analyzer (model 42i Trace Level Nitrogen Oxide Analyzer, Thermo Fisher Scientific, USA) on chemiluminescence, the O3 analyzer (model 49i, Thermo Fisher Scientific, USA) on UV absorption, the SO2 analyzer, (model 43i, Thermo Fisher Scientific, USA) on UV fluorescence, and the PM2.5 analyzer (model SHARP 5030, Thermo Fisher Scientific, USA) on light scattering and beta attenuation. Meteorological parameters, including air temperature, pressure, relative humidity, and wind speed and direction, were monitored using a WXT520 weather station (Vaisala, Finland).

GEM-MACH is a chemical transport model embedded within the GEM (Global Environmental Multiscale) numerical weather forecast model of Environment and Climate Change Canada (ECCC) (Côté et al., 1998a, b). Meteorological conditions (Makar et al., 2015b) and air quality processes, including gas-phase, aqueous-phase, and heterogeneous chemistry and size-resolved aerosol processes, are included in GEM-MACH (Moran et al., 2010; Gong et al., 2015; Makar et al., 2015a). GEM-MACH is used operationally by ECCC for short-term air quality forecasting on a North American grid with 10 km horizontal grid spacing (Moran et al., 2014; Pavlovic et al., 2016). In this study, a research version of GEM-MACH simulated concentrations of pollutants with a horizontal grid-cell size of 2.5 km within a 40 m layer above ground level. Our measurement site was located within one model grid cell. Hourly outputs were obtained from GEM-MACH in this study.

GEM-MACH outputs instantaneous pollutant mixing-ratio fields once an hour, including CO, O3, NH3, HCHO, NO, and NO2. The FTIR spectrometer and the NAPS measured mixing ratios once a minute. In order to compare model results and measurements for similar periods, GEM-MACH results were averaged over the two bracketing timestamps to get an estimate of the average mixing ratio of pollutants over each hour, while the measurements results were averaged every hour to match the temporal resolution of GEM-MACH results.

Various approaches have been developed to deduce source emission rates from pollutant concentrations, including inverse dispersion models (see Flesch et al., 2004). We used a backward Lagrangian stochastic (bLS) model (WindTrax, http://www.thunderbeachscientific.com; Flesch et al., 1995) to calculate the emission rate Q through Q=C-Cb(C/Q)sim, where C is the concentration of a pollutant at the measurement site, Cb is the background concentration, and (C/Q)sim is the simulated ratio of concentration at the site to the emission rate upwind. In the bLS model, a large number of virtual particles is released at the site, and individual upwind trajectories are calculated backward in time from the site. Then the fraction of trajectories that originate from the user-predetermined source area is calculated, which in turn factors into the calculation of (C/Q)sim. WindTrax can handle complex source-area shapes but not variations in topography. The micrometeorological condition inputs for the bLS model are u∗ (friction velocity) and L (Obukhov length) obtained from the scintillometer measurements (Supplement Sect. S2) as well as wind and ambient temperature data from the NAPS trailer.

In this study, we used the mixing ratio of CO retrieved from FTIR spectroscopy to estimate the CO emission rate from the highway (a “top-down” approach). The estimated emission rates are then compared to the emission rates derived from traffic volumes combined with published emission factors of vehicle engines, i.e., a “bottom-up” approach (Sect. 3.7). These results will help evaluate the capability of deducing emission rates from our measurements.

Traffic volume data were provided by the Ontario Ministry of Transport, in units of vehicles per hour passing a point on Highway 401. Before 20 July, counts were available at the Islington Avenue intersection, about 700 m to the west of our site. However, after 20 July, data at Islington Avenue were not available, and we instead used traffic volume data at a nearby intersection to the east of our site at Avenue Road, which showed a linear relationship with traffic volumes at Islington Avenue. Therefore, the traffic volume data before 20 July were measured at the Islington Avenue intersection, while data after 20 July for the same location were estimated.

During the study, the mean wind speed measured at the NAPS trailer was 2.5 m s-1, with a range from 0 to 9.9 m s-1 and quartiles of 1.3 and 3.3 m s-1. The mean friction velocity u∗ during the study was 0.40 m s-1, with a range from 0.02 to 1.31 m s-1 and quartiles of 0.25 and 0.52 m s-1. The mean ambient temperature was 24∘C, with a range from 14 to 33 ∘C.

In Toronto in late July, sunrise occurs at about 06:00 EDT (eastern daylight time; same time labels were used for the entire study), solar noon occurs at about 13:30, and sunset occurs at about 21:00. As shown in Fig. 2, sensible heat flux H started to increase beginning at 06:00, reached its maximum in the early afternoon around 13:30, and then decreased to its minimum after 23:00. The downwelling shortwave solar radiation started to increase at 06:30 and reached a peak around 13:00. It is notable that H remained positive throughout the night and started to increase before sunrise. We surmise that this is due to traffic providing a source of sensible heat and mechanical and convective turbulence, as well as slow release of heat from the pavement at night (Sailor and Lu, 2004; Khalifa et al., 2016). A rough estimation of 33 W m-2 (56 % of H) at 05:30 (before the sunrise) was contributed by vehicles on the highway, based on the traffic volume, the ratio of energy loss from gasoline engine, and the typical fuel consumption of gasoline vehicles. Results of z/L remained negative throughout the night, indicating that the surface layer was always unstable or neutral. u∗ also varied diurnally, with higher values from 08:00 to 21:00 and lower values during the night, which also suggests stronger turbulence in the daytime and is correlated with H. All of these measurements show that mixing and turbulence started to increase quickly after sunrise, reached a maximum in the early afternoon, and decreased to a minimum after 23:00.

NH3 can form secondary aerosols that are associated with negative health effects (Seinfeld and Pandis, 2006; Behera and Sharma, 2012; Liu et al., 2015) as well as radiative forcing impacts. According to the trend data for 2016 of the US Environmental Protection Agency (U.S. EPA, 2016), 2.4 % of US national NH3 emissions are from vehicles which are more important sources in urban regions. After the three-way catalytic converter (TWC) was introduced to gasoline vehicles in 1981 and became used widely, NH3 (as a product formed in TWC from the reaction of NO with CO and H2O) emissions from vehicles increased (Moeckli et al., 1996; Fraser and Cass, 1998; Kean et al., 2000). NH3 is also involved as a reagent in the reduction processes for NO in selective catalytic reduction converters (SCR) in diesel vehicles. Therefore, diesel vehicles could also contribute to NH3 emissions due to the aging of catalysts and over-doping of urea. However, they play only a minor role in NH3 traffic emissions compared to gasoline vehicles based on the emission inventory used by GEM-MACH over the Greater Toronto and Hamilton Area (projected 2015 inventory based on air quality modelling version of 2010 Canadian Air Pollutant Emission Inventory, Environment and Climate Change Canada, Ottawa, Ontario, November 2014, unpublished). NH3 is gaining importance as a pollutant from traffic due to the gaining use of emission control systems, but previous studies which directly measured NH3 mixing ratio from traffic are rare. Elevated mixing ratios of NH3 between 0 and 23 ppb were observed with the FTIR spectrometer in this study (Fig. 6). Baldauf et al. (2008) showed diurnal plots for traffic volume and mixing ratios of NH3 retrieved from open-path FTIR spectroscopy 20 and 300 m from a main road. The NH3 mixing ratio they reported was between 10 and 35 ppb, comparable to our results.

Traffic emissions appear to be very important to NH3 in urban environments, although residential garbage collection (Reche et al., 2012), soil and fertilizers, biomass burning, natural ecosystems, sewage and landfill, and direct emissions by humans and animals could also contribute (Sutton et al., 2000). Yao et al. (2013) found a good linear correlation between mixing ratios of NH3 and NO during periods in the morning at the same site beside Highway 401. CO has been used as a common reference pollutant from vehicle emissions as discussed in Sect. 3.2.1, and studies have also used [NH3] / [CO] ratio to correlate NH3 to traffic emissions (Perrino et al., 2002; Livingston et al., 2009). A linear correlation between emission factors of CO and NH3 was found in light- and medium-duty vehicles in the California South Coast Air Basin by Livingston et al. (2009). Perrino et al. (2002) found a linear relationship between mixing ratios of NH3 and CO at a traffic site in Rome. A linear relationship between NH3 and CO mixing ratios from the FTIR spectrometer over the whole period was also observed in this study (slope = 0.023, r2=0.60, Fig. S5). This linear relationship suggests that NH3 and CO shared a common source; in this case, a significant fraction (76 % during morning rush hour; see discussion in the next paragraph) of NH3 came from traffic. The slope of 0.023 ([NH3] / [CO]) is close to values previously reported (Livingston et al., 2009). NH3 emission factors from vehicles in the literature are in the range of 0 to 0.144 g km-1 depending on various factors such as fuel type, driving cycle, vehicle engine power, engine temperature, and catalyst aging (Durbin et al., 2002; Huai et al., 2003). Therefore, differences in slopes among studies are to be expected.

Average diurnal cycles of NH3 mixing ratios on weekdays and weekends are shown in Fig. 5. To estimate the NH3 due to traffic emissions, it was assumed that traffic emission was the only source of CO above background at this spatial scale and all NH3 from traffic emissions were correlated with CO. Thus, a background CO mixing ratio of 265 ppb was subtracted from the retrieved CO mixing ratio and the result was regressed against NH3 mixing ratios, resulting in a traffic-related NH3 being estimated as 0.023 × ([CO]_FTIR-265) ppb. The 265 ppb CO background was the intercept of CO from the linear regression of NH3 with CO. The resulting weekday and weekend diurnal cycles of the estimated NH3 mixing ratio due to traffic emissions are plotted in Fig. 5. During the morning rush hour and late at night on weekdays, traffic emissions contributed more to NH3 levels than during other times of day. On weekends, the diurnal variation of total NH3 was weaker, and estimated NH3 from traffic accounted for essentially all NH3 observed. Overall, there is no indication of a background offset of NH3, and most measured NH3 at this site can be accounted for by traffic emissions.

NH3 mixing ratio retrieved from FTIR spectroscopy also agreed well with GEM-MACH model simulations (Fig. 6). The analysis results of the traffic contribution to NH3 around the site based on the FTIR measurements are consistent with the GEM-MACH model NH3 input emissions, which show that the main source of NH3 at this location is vehicular (Fig. S3).

O3 is a secondary pollutant and is not emitted directly by vehicles. NO reacts with O3 forming NO2 on a timescale of a few minutes during the day. Photochemistry between VOCs and ambient oxidants produces O3, and HCHO is one of the products from these photochemical reactions. The chemistry of titration and photochemical production of O3 has been discussed previously in detail (Marr and Harley, 2002b; Fujita et al., 2003; Seinfeld and Pandis, 2006; Murphy et al., 2007). The time series of O3 mixing ratio from the FTIR spectrometer also agrees broadly with the NAPS O3 measurements (Fig. 3). However, the polar plot in Fig. 7a shows that O3 mixing ratios measured by the FTIR spectrometer and the NAPS were close when the wind was from the highway, whereas O3 from the FTIR spectrometer was much lower than O3 from the NAPS, when the wind was from other directions. These results can be explained by the titration of O3 over the highway by NO emissions from vehicles: when the wind is from the north, the O3 reaching the NAPS trailer has been titrated, but when the wind is from the south, O3 measured at the NAPS site is titrated over the highway downwind of the measurement point.

Mixing ratios of NO and NO2 can be retrieved from the FTIR spectra, but the correlation coefficients of fitting are less than 0.1 and estimated mixing ratios contain large offsets and biases, probably due to the strong interference from water vapour. Therefore, mixing ratios of NO and NO2 from the FTIR spectrometer are not shown here. The GEM-MACH simulations and NAPS measurements for NO and NO2 often do not agree well (Fig. 8). The disagreements can again be partially explained by the influence of wind direction. Like CO and NH3, NO is directly emitted from vehicles, but it reacts in the atmosphere much more quickly than CO or NH3. Polar plots for NO and NO2 (Fig. 7b and c) show the effect of wind direction on the mixing-ratio differences between GEM-MACH results and NAPS measurements. When the wind blew from the NAPS trailer towards the highway, the difference was close to zero, but when the wind blew across the highway towards the trailer, GEM-MACH predictions were significantly lower than NAPS measurements. Similar to the CO comparison, the NAPS measurements were strongly influenced by traffic emissions when the wind came from highway compared to GEM-MACH. Note that GEM-MACH simulates mean pollutant mixing ratios within a 40 m layer and the inlet of the NAPS trailer was about 3 m above the ground. These different heights also contribute to the disagreement between measurements and model results. The NOx (NOx = NO + NO2) mixing ratio measured at the NAPS station on weekdays showed a similar average diurnal cycle (Fig. 9) to CO by the FTIR spectrometer, reaching a peak over 100 ppb from 06:00 to 08:00 followed by significant decrease in the middle of the day and a secondary peak between 20:00 and 23:00. The diurnal cycle of NOx on weekends with mixing ratios of 0–35 ppb over the whole day, significantly lower than weekday NOx levels, was also less variable. Reduced NOx levels on weekends may have been due to fewer diesel vehicles operating on weekends; this pattern has been reported in studies in California (Marr and Harley, 2002a; Harley et al., 2005; Kim et al., 2016). Zhang et al. (2012) found that fewer diesel vehicles were observed on weekends on another major highway in the Toronto area. The annual sales of fuel used for on-road motor vehicles in Canada in 2015 were 42.6 billion L of gasoline and 18.0 billion L of diesel (Statistics Canada, 2016); i.e., a significant fraction of fuel burned is diesel. Therefore, lower diesel vehicle volumes on weekends may have contributed to different emissions of NOx on Highway 401 near our site. The [NOx] / [CO] ratio also has been used to check the chemical conditions related to O3 production. Figure 10 shows that the ratios in this study are 0.20 and 0.10 for weekdays and weekends, respectively. The lower ratio on weekends is likely due to reduced numbers of diesel vehicles, which is consistent with a previous study by Kim et al. (2016). Our [NOx] / [CO] ratios during both weekdays and weekends are greater than their results (0.11 and 0.033), but our study only focused on near-surface observations over a short defined section of Highway 401 while their observations were at 1 km above the ground level with a bigger footprint which included off-road emissions and other local sources. Hassler et al. (2016) showed the trend of [NOx] / [CO] in the Los Angeles Basin, and the ratio is between 0.1 and 0.2 after 2010, which agrees well with our results. There are also some previous studies showing ratio of [CO] / [NOx] from regions near heavy traffic emission. Parrish et al. (2012) reported the slope of [CO] vs. [NOx] was in the range of 6.3 to 18.9 for the measurement from 1987 to 1999. Wallace et al. (2012) reported the slope of [CO] vs. [NOx] was 4.2 in the morning rush hours in 2009. The slopes of [CO] vs. [NOx] in this study are 3.14 and 7.75 for weekdays and weekends, respectively. Therefore our results on NOx and CO are comparable with these previous studies.

Figure 9 also shows average weekday and weekend diurnal cycles for O3 measured at the NAPS station. One interesting feature is that the median diurnal O3 mixing ratios on weekends were consistently greater than on weekdays. Also, the diurnal cycles of O3 were inversely correlated with those for NOx .The low O3 mixing ratios in the mornings of weekdays can be explained by titration with high fresh emissions of NO from traffic, whereas the afternoon maximum is mainly due to production of O3 through increased levels of photochemistry with VOCs. The diurnal cycle of odd oxygen (Ox = O3 + NO2) shown in Fig. 9 can be used to separate the contributions of titration and photochemistry with VOCs to O3 mixing ratios. Titration does not increase the sum of O3 and NO2, whereas photochemistry with VOCs does. Therefore, variations in Ox levels indicate that photochemistry with VOCs is important (Pollack et al., 2012). The average diurnal mixing ratios of O3 and Ox from NAPS measurements showing a maximum in the afternoon and being slightly higher on the weekends is also consistent with the average diurnal mixing ratios of HCHO showing a peak in the afternoon on weekends (Fig. 11). These diurnal results also suggest that the photochemistry with VOCs producing NO2 and O3 was important especially in the afternoon and on weekends. In addition, Ox levels peaked in the afternoon, also consistent with diurnal cycles of sunlight intensity (Fig. 2), which is a critical condition of photochemistry to produce O3. Other VOCs besides HCHO that were emitted by traffic and other local sources may also have contributed to the photochemical production of O3, but they were not quantified. Similar differences of O3 and Ox mixing ratios between weekdays and weekends were reported in the South Coast Air Basin (Pollack et al., 2012; Warneke et al., 2013). Temperature also affects O3 production, but given the great variation of O3 mixing ratio through the day, this was a secondary effect here based on box model calculations in the temperature range of our study (Coates et al., 2016).

To evaluate how representative the contrasts between weekends and weekdays based on this 16-day data set are compared to longer time frames, three summers of O3 measurements from a nearby NAPS station were extracted and analyzed (Supplement Sect. S3 and Fig. S7). Similar diurnal patterns and differences were observed in 2 of the 3 years, suggesting that the analysis presented above is representative of longer terms as well.

Time series of HCHO mixing ratios retrieved from the FTIR spectrometer shown in Fig. 6 were between 0 and 5 ppb. The average diurnal cycle of HCHO during weekends reached a peak in the early afternoon (shown in Fig. 11), which corresponded not to the average diurnal cycles of either traffic volume or primary pollutant CO but rather to the sunlight intensity (i.e., actinic radiation; see Fig. 2). This indicates that photochemistry of VOCs with oxidants was a dominant source (Stroud et al., 2016). The lifetime of HCHO in the atmosphere is on the scale of hours to days depending on the levels of ambient oxidants (Seinfeld and Pandis, 2006). Stroud et al. (2016) reported on levels of HCHO in Toronto and Egbert, Ontario, and source apportionment. Primary mobile emissions were found to contribute ∼ 12 % of HCHO in Toronto. Previous research also showed that both light-duty and heavy-duty vehicles emit HCHO (Grosjean et al., 2001).

The correlation between HCHO mixing ratio and ambient temperature was moderate (Fig. S6a, r2=0.42). HCHO levels were low for a few weekdays (22–24 July) with lower temperature as compared to the four warmer days sampled on the weekends. Therefore, the difference between the average weekday and weekend diurnal cycles shown in Fig. 11 may be due in part to sample size, and it is possible that other local HCHO emission sources which depend on the temperature may also have contributed to the HCHO observed, especially in the afternoons on weekends.

GEM-MACH simulations of HCHO mixing ratio are always greater than the FTIR measurements, but in the GEM-MACH HCHO model species from the ADOM-II gas-phase chemistry mechanism is actually a lumped species that also includes isoprene oxidation products. Therefore, the GEM-MACH results of HCHO are not shown here.

The mixing ratios of HCN retrieved from FTIR measurements were between 0 and 4 ppb (Fig. 6). Only on 28, 29, and 30 July was the HCN observed above its detection limit. HCN has severe adverse effects on human health, and chronic exposure to low cyanide can cause abnormal thyroid function and neurological problems (El Ghawabi et al., 1975; Blanc et al., 1985; Banerjee et al., 1997; U.S. EPA, 2010b). HCN has been reported previously in vehicle exhaust (Bradow and Stump, 1977; Keirns and Holt, 1978; Cadle et al., 1979; Urban and Garbe, 1979, 1980; Karlsson, 2004; Baum et al., 2007; Moussa et al., 2016). It may form over the catalytic converters in the vehicle emission control systems (Voorhoeve et al., 1975; Suárez and Löffler, 1986; Baum et al., 2007). A recent study for Toronto reported comparable HCN mixing-ratio values (Moussa et al., 2016). These HCN measurements contribute to the few studies reporting measurements of HCN mixing ratio beside a highway in urban ambient air.

As shown in Fig. 6, mixing ratios of CH3OH from the FTIR spectrometer were between 2 and 20 ppb most of the time, with some high spikes. Figure 12 presents the corresponding average weekday and weekend diurnal cycles of CH3OH for the study period. This plot shows the mixing ratio reached a peak (maximum of 20 ppb at 07:30) from 07:00 to 09:00 on weekdays whereas there was no peak in the mornings on weekends. In addition, a linear relationship between [CH3OH] and [CO] was observed during the early morning rush hours on some weekdays (Table 2). These results suggest that at least a fraction of observed CH3OH was from traffic emissions. Observations of methanol associated with traffic have been reported in other studies. Rogers et al. (2006) reported CH3OH in the diluted pipeline exhaust of a mobile laboratory. CH3OH may also come from non-engine sources, such as windshield wiper fluid. Durant et al. (2010) measured gas and particle pollutants near Interstate 93 in Massachusetts. They reported CH3OH was above 20 ppb at 07:20 50 m downwind of the highway, possibly with contributions from some other local sources. Reyes et al. (2006) reported vehicle emission of non-regulated pollutants, including methanol, by using local gasoline and driving conditions in Mexico City. Rantala et al. (2016) studied urban VOC fluxes in urban Helsinki and found methanol fluxes were correlated with traffic and with CO fluxes, traffic could partially explain the observed methanol. Sahu and Saxena (2015) also reported CH3OH mixing ratios at Ahmedabad (an urban site in India), and both traffic emission and the transport from biomass burning and biogenic sources outside the city contributed to CH3OH. The mixing ratio of CH3OH we observed did not correlate with ambient temperature (Fig. S6c), so there was no strong indication of biogenic sources.

To evaluate the feasibility of using measurement data from this study to estimate emission rates, we picked measurements for three days (22, 28, and 29 July) from this study to use as inputs to a backward Lagrangian stochastic dispersion model (WindTrax, http://www.thunderbeachscientific.com/). 22 July was chosen, because the wind direction was steadily from northwest and a traffic jam occurred for added interest. 28 and 29 July were chosen, because they are two of the highest days for temperature and O3 during this project. The following were included as the inputs: CO mixing ratio from the FTIR spectrometer; background mixing ratio of CO; winds and temperature from the NAPS trailer; and atmospheric stability (u∗ and L) from the scintillometer. The surface roughness (z0) was set to 0.15 m, the maximum allowed by WindTrax. The defined section of Highway 401, which was assumed to be the only CO source in the footprint, is about 1870 m long and 110 m wide. The FTIR path is roughly in the centre of the defined section. WindTrax was then used to estimate the emission rate of CO from this defined section. In the model, 50 000 virtual particle trajectories were calculated upwind of the FTIR path with the given meteorological conditions (wind direction and temperature) and surface-layer turbulence (u∗ and L), to determine what fraction of trajectories originated from the designated source area.

Over 3 days, 22, 28, and 29 July, CO emission rate estimates (g h-1 m-2) were calculated by WindTrax with a 1-minute resolution for 10-minute periods, and the average estimates over those 10 min periods were calculated and shown as the markers in Fig. 13. The constant background used in WindTrax was 265 ppb, which was determined from the CO intercept of the linear regression analysis of NH3 with CO (see Sect. 3.3). The mixing ratio of changing background used in WindTrax was determined using a more dynamic definition of background based on wind direction. When the wind was from the south, the background was chosen as the NAPS measurement for 28 and 29 July. In the morning on 28 July when the wind was from the northwest, the background was chosen as 415 ppb, the minimum mixing ratio of that morning measured by the FTIR spectrometer. When the wind direction varied greatly, the background value of the previous hour was chosen. On 22 July, the wind was consistently from the northwest, and the minimum of 329 ppb over the whole day from the FTIR spectrometer was chosen as the changing background.

Liu and Frey (2015) reported vehicular empirical cycle average emission factor for specific pollutant, vehicle, and driving cycle in grams per mile as well as average, minimum, and maximum values, based on empirical data measured between 2008 and 2013 in the Raleigh and Research Triangle Park area (North Carolina, USA) for 100 vehicles with a range of model years and accumulated mileage. We converted these results into g km-1 emission factors ranging from 0.003 to 5.1 g km-1 with an average value of 0.62 g km-1. They also reported simulated CO emission factors ranging from 0.004 to 6.87 g km-1 with an average value of 1.99 g km-1 by using the EPA's Motor Vehicle Emission Simulator (U.S. EPA, 2010a) emission factor model. Moussa et al. (2016) reported that emission factors of CO measured from several gasoline light-duty vehicles with different driving cycles ranged from 0.1 to 3.0 g km-1 with an average value of 0.9 g km-1.

Using emission factors from the MOVES model, traffic volume estimates, and the width of Highway 401 at the site, a “bottom-up” estimate of the emission rate (g h-1 m-2) was calculated and compared to the dispersion model results (Fig. 13). There is a good agreement amongst the different approaches (-140 to 460 and -110 to 70 %; see Fig. 13 and discussion below). On 22 July, the emission rate estimates from the WindTrax are close to the MOVES results. The MOVES estimates show a sharp drop between 14:00 and 15:30 due to decrease of traffic volume to 14 % of at the value at 13:30. During this time, the emission rate estimates from the WindTrax do not fluctuate much. This result suggests that vehicle numbers passing by a fixed point may not be the best indicator of emissions since they do not account for the traffic speed: an extreme example would be a traffic jam with zero traffic flow but nonzero emissions. On 28 and 29 July, the estimates from the WindTrax are greater than the MOVES estimates. The difference between the emission rates estimated by using the WindTrax with changing background and by using 0.9 and 3.0 g km-1 from Moussa et al. (2016) is in the range of -140 to 460 and -110 to 70 %, respectively. These results suggest that WindTrax dispersion calculation results based on CO mixing-ratio measurements from the FTIR spectrometer and in situ micrometeorology are well within the range of estimates based on traffic volume and emission factors of various vehicles.

The input background mixing ratio assumed at each hour influences the emission rate estimates. Especially during the night around 00:00 to 06:00 when the wind was from north, it is difficult to determine the background mixing ratios of CO, since no measurements were available upwind. Emission rate estimates with both constant background and changing background were calculated to check the sensitivity of background mixing ratio on the emission rate estimates by the WindTrax. Our changing-background approach should be closer to reality, since conditions around the highway do change with time. Figure 13 shows that both estimates from the WindTrax using changing background and constant background agree in general with the bottom-up estimates, except for the period of the morning on 29 July when WindTrax estimates with the constant background are greater than the other estimates. The wind was from the south during this period and the assumed constant background of 265 ppb was significantly lower than the NAPS measurements, resulting in an overestimate by the WindTrax. Beside the uncertainties in background mixing ratio, the variations in emission rate estimates are most likely due to changes in wind direction over short periods. When the wind direction changed quickly, the input wind direction used by the WindTrax may not be representative and hence may bias the calculated emission rate.

Emission rates of other primary pollutants from traffic can be determined by using the concentration ratios of these pollutants to CO and emission rate estimates of CO, as mentioned in Sect. 3.2.1. We found that [NH3] (as discussed in Sect. 3.3), [CH3OH], and early morning periods of [NO] had linear relationships with [CO]. With these ratios and emission rate estimates of CO obtained from the WindTrax (Fig. 13), the 10-minute average emission rate estimates of NH3, NO, and CH3OH were calculated. The minimum-to-maximum range and the average of these 10-minute average estimates are shown in Table 2. Emission factors in g km-1 were calculated from WindTrax emission rates estimates by using the 110 m width of that section of highway and traffic volume to compare our estimates to previously reported emission factors.

The inputs of u∗ and L from the scintillometer measurements also contain uncertainties. A realistic estimate of the uncertainty of u∗ is ±20 to 30 % (Andreas, 1992). We conducted a sensitivity study by varying u∗ from 0.7 to 1.3 u∗(obs) and L from 2.20 to 0.34 L(obs) corresponding to change of u∗ while keeping heat flux fixed, which resulted in emission rates from 0.69 to 1.37 times the original emission rate estimates. We also investigated the sensitivity of heat flux on emission rates estimates by varying L from 0.5 to 2L(obs) with fixed u∗(obs), which resulted in emission rates estimates in the range of 0.70 to 1.45 times the original emission rate estimates. Therefore, even with conservative uncertainty estimates about surface-layer stability, the calculated emission rates from the WindTrax are still within 45 % of the bottom-up estimates.

WindTrax limits z0 to a maximum value of 0.15 m. However, the z0 of the actual measurement site over the highway is around 0.6 m based on urban-scale meteorological model results for Toronto (Leroyer et al., 2016). This difference between the actual z0 and the input z0 used by WindTrax likely also contributes to the uncertainty of the emission rate estimates.

Better emission rate estimates might also be obtained if traffic information included vehicle types and vehicle speed. Speed and speed variation of different vehicle types are known to affect emission rates of CO, hydrocarbon, and NOx (Zhang et al., 2011), but consideration of these factors would have required more sophisticated traffic quantification.