Air quality and meteorological measurements were made from Environment and Climate Change Canada's mobile laboratory: Canadian Regional and Urban Investigation System for Environmental Research (CRUISER) (Levy et al., 2014; Government of Canada, 2018). CRUISER was deployed during two seasons over the Greater Toronto Area (GTA), a metropolitan area encompassing the city of Toronto and four regional municipalities with a population of over 6 million. The summer campaign took place over 9 days in July 2015 (15, 16, 17, 20, 21, 22, 23, 27, and 28 July) as part of the Environment Canada Pan and Parapan American Science Showcase (ECPASS) (Joe et al., 2018). The winter campaign took place over 8 days in January 2016 (11, 13, 14, 15, 18, 19, 20, and 21 January) as part of a health exposure mapping study. Driving took place on weekdays only, with the majority of measurements occurring between 09:00 and 17:00 LT (local time). Driving routes were chosen to pass along highways, major roadways, and local streets, and to visit residential, commercial, and industrial areas; the driving routes for the summer and winter campaign are shown in Fig. S1 in the Supplement. In 2016 the Ontario vehicle fleet was composed of approx. 97 % light-duty vehicles (LDVs; vehicles < 4500 kg and motorcycles/mopeds) and 4 % heavy-duty vehicles (HDVs; for this paper, the HDV category includes both medium-duty vehicles 4500–14 999 kg and heavy-duty trucks > 15 000 kg, as well as buses) (Statistics Canada, 2016); the composition of the GTA vehicle fleet is assumed to be similar.

Several gas-phase and particle-phase instruments were housed on-board CRUISER as listed in Table 1. Carbon dioxide (CO2) was measured with 2 s time resolution by cavity-enhanced laser absorption spectroscopy (PICARRO). All gas-phase instruments sampled from a common inlet with the exception of the high-resolution time-of-flight chemical ionization mass spectrometer (HR-TOF-CIMS) which sampled off a dedicated inlet located on the roof of CRUISER towards the rear right side. The common gas-phase inlet was located 3.6 m a.g.l. and oriented near the front left side. Ambient air was sampled through a 2 m long PFA tube with 0.61 cm ID followed by a 30 cm long PFA tube with 0.52 cm ID at a rate of 13.6 SLPM (residence time ∼0.25 s); inlet lines for various instruments were connected downstream of this common inlet. All particle-phase instruments sampled off a common stainless steel inlet located adjacent to the gas-phase inlet. During the winter campaign, CO2 was sampled from the same inlet as the HR-TOF-CIMS. Relative wind speed and wind direction was measured using an ultrasonic anemometer located on the roof at the front of CRUISER. Periods of potential self-sampling were identified and removed using an algorithm which is described in the Supplement. Briefly, the self-sampling algorithm identified periods of potential exhaust based on CRUISER speed, relative wind speed, and wind direction (towards inlet) and periods of suspected exhaust within these windows, based on the presence of exhaust tracers (BC, NO, fine particle counts). Periods of suspected exhaust were removed from the data.

Volatile organic compounds were measured using a proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF-8000, Ionicon Analytik). The operating principles of the PTR-TOF-MS instrument have been described elsewhere (Jordan et al., 2009; Li et al., 2017); further details can be found in the Supplement. Briefly, the PTR-TOF-MS was operated with an E / N value of 140 Td. Air for analysis by the PTR-TOF-MS was sampled off the common gas-phase inlet via a 2 m long PFA tube with 0.52 cm ID at a rate of 4.4 SLPM and the instrument sampled part of this flow (100 sccm) through a 120 cm insulated PEEK capillary with 0.08 cm ID heated to 70 ∘C. Mass spectra were acquired with a time resolution of 1 s and a resulting mass resolution of approx. 4000 m Δm-1. The response of the PTR-TOF-MS to specific VOCs was determined using a home-built zero and calibration unit and a custom VOC gas standard (Ionicon). The 2σ detection limits differed slightly for the summer and winter campaigns and were calculated respectively to be 110 and 155 pptv for benzene, 125 and 240 pptv for toluene, and 110 and 160 pptv for C8 benzenes. The sensitivities and detection limits are also listed in Table S1 in the Supplement.

HNCO and HCN were measured using a high-resolution time-of-flight chemical ionization mass spectrometer (HR-TOF-CIMS, Aerodyne Research, Inc.). The design, operation, and mobile deployment of the HR-TOF-CIMS has been previously described (Veres et al., 2008; Roberts et al., 2011; Wentzell et al., 2013; Liggio et al., 2017a). Additional details can be found in the Supplement. Briefly, the HR-TOF-CIMS is a differentially pumped time-of-flight mass spectrometer configured to use iodide ion as the reagent ion (Woodward-Massey et al., 2014; Le Breton et al., 2013). Air for analysis was drawn at ∼22 SLPM through a 3 m long heated (50 ∘C) inlet (0.58 cm ID). The CIMS subsampled from this flow into the ion molecule reaction (IMR) region via a critical orifice at 1.7 SLPM. Mass spectra were acquired with a time resolution of 1 s and a resulting mass resolution of approx. 5000 m Δm-1. Calibrations of HNCO were conducted by thermally decomposing cyanuric acid at 250 ∘C to HNCO (Roberts et al., 2010) with the permeation rate quantified via Fourier-transform infrared spectroscopy (FTIR; Thermo-Fisher Inc.). Calibrations of HCN were performed by diluting a HCN gas standard (Air Liquide, ppmv in N2) in zero air. Humidity-dependent response factors for both species were derived by diluting the calibration gas flows with humidified air to a final RH ranging from ∼9 % to 90 %, resulting in sensitivities of 0.086 and 0.1 ncps pptv-1 for HCN and HNCO respectively. The 2σ detection limits for HNCO and HCN were estimated to be 7 pptv each for both the summer and winter campaigns.

Black carbon measurements were made with a high-sensitivity laser-induced incandescence (HS-LII) instrument (Atrium Technologies Inc., CA, USA) developed in collaboration with the National Research Council Canada (NRC). The particular instrument on CRUISER is a research-grade prototype capable of ultra-low BC measurements at 1 s resolution. Here, black carbon is operationally defined by its high thermal stability (Petzold et al., 2013). The principle of operation of this instrument, as well as its use during ambient studies, has been described elsewhere (Snelling et al., 2005; Chan et al., 2011; Liggio et al., 2012). Briefly, ambient particles within a set volume are rapidly heated by a pulsed laser beam (1064 nm; 7 ns FWHM, 200 mJ pulse-1) to just below the soot sublimation temperature (∼4000 K). The absolute incandescence and temperature of the BC particles are measured by collection optics and photomultipliers. After appropriate calibration and analysis, these two parameters are used to determine the soot volume fraction, which is converted to a BC mass concentration with knowledge of the particle material density (ρ) and the absorption function (Em), both of which are well established for BC (Coderre et al., 2011; Choi et al., 1994; Wu et al., 1997). An advantage of this technique is that it determines ensemble properties for all particles within the sample volume and so does not suffer from a particle size limitation; previous studies have shown that the HS-LII can detect laboratory-generated particles < 7 nm in diameter (Stirn et al., 2009). As a result, a previous study found that BC measurements by a single-particle soot photometer (SP2), which is only sensitive to particles with a diameter > 70 nm, are biased low relative to the HS-LII (Liggio et al., 2012). Furthermore, it has been shown that the HS-LII is significantly less influenced by the presence of non-refractory mass compared to other BC measurement methods such as photoacoustic spectrometers (Chan et al., 2011). The HS-LII was only in operation for the summer campaign.

Pollutant and CO2 time series were averaged to 2 s and then further smoothed using a 3-point boxcar (5 s). The background (BKG) was subsequently defined as the rolling second percentile over a 90-point (180 s) window, with additional boxcar smoothing over the same window. Similar approaches for estimating background concentrations from mobile monitoring studies have been employed by others (Jiang et al., 2005; Jimenez et al., 2000; Hudda et al., 2013; Park et al., 2011; Bukowiecki et al., 2002; Larson et al., 2017). Since the background is calculated over a 3 min window, corresponding to approximately 2 km of CRUISER travel, it is assumed to be representative of a neighborhood-scale background (Larson et al., 2017). The on-road or LOCAL concentrations are defined as the background-corrected (i.e., above-background) mixing ratios. Figure S2 shows sample time series for the summer (CO2, benzene, BC, HNCO, HCN) and winter campaigns (CO2, benzene, HNCO, HCN) and demonstrates that the LOCAL pollutant plumes frequently co-varied with increases in CO2, suggesting a combustion (i.e., vehicular) source.

An emission factor algorithm was written using Igor Pro (Wavemetrics Inc.) to identify CO2 plumes based on the first and second derivatives of the CO2 time series, similar to the approach of Wang et al. (2015). The details of the algorithm can be found in the Supplement. Briefly, the first derivative of the CO2 time series was used to identify peak boundaries and locations (peak maxima). Two types of plumes were identified: single-peak plumes (SPP) and multi-peak plumes (MPP). Multi-peak plumes contain one or more CO2 peaks (and include the SPP set). Plumes less than 10 s in duration or with an average background-corrected CO2 response of <5 ppmv s-1 over the integration period were rejected as erroneous or uncaptured (Wang et al., 2015). Emission factors (EF) expressed as mg kgfuel-1 for a given pollutant X and plume i were calculated using a carbon mass balance approach: EFX,i=[X]CO2×MWXFC⋅MWC×Cfuel×103, where [X] and [CO2] are the integrated amounts of LOCAL (background-corrected) X and CO2 over the boundaries of plume i in units of ppbv and ppmv respectively, MWX and MWC are the molecular weights of pollutant X and carbon in g mol-1, FC is the molar ratio of carbon in CO2, Cfuel is the carbon mass fraction in the fuel in kg C kgfuel-1, and 103 is the necessary unit conversion factor. A value of Cfuel=0.86 was used here, which is the average of the Cfuel for gasoline (0.85) and diesel (0.87) (Wang et al., 2015). Strictly, the denominator in Eq. (1) should contain the sum of all emitted carbon species (CO2, CO, total hydrocarbons); however, emissions of CO2 have been shown to account for >90 % of fuel consumption (Jathar et al., 2017; Yli-Tuomi et al., 2005). Plumes associated with the highest EFs were visually inspected and in some instances were deemed to have been erroneously captured based on poor correlation between the pollutant and CO2 time series; these plumes were removed from the final dataset. Further details regarding the background calculation and their influence on calculated EFs, as well as the peak removal processes, can be found in the Supplement (Sect. S1.5).

Emission factors for benzene, toluene, C2 benzenes, NO, NO2, NOx, PN, and BC were obtained during the summer campaign. EFs for HCN and HNCO were obtained during the winter campaign, when the PICARRO measuring CO2 shared the same inlet as the HR-TOF-CIMS. Statistics on the number of plumes, plume duration, and number of peaks per plume can be found in Tables S2 and S3 at various stages of analysis for the summer and winter campaigns. We note that for NO, reactions with ozone can result in a low bias for NO EFs. In this study we expect the time from emission to be on the order of minutes, although exact emission times are not known. As such it is likely that the EFs for NO here represent lower limits to the true NO EFs.

Emission factors were also calculated using the time-based approach, which considers the entire data set, in contrast to the plume-based approach which only considers periods of elevated CO2 as defined by peaks (Westerdahl et al., 2009). The LOCAL (background-corrected) CO2 and pollutant mixing ratios were integrated in consecutive intervals of 30, 60, 90, and 120 s and fuel-based emission factors were calculated according to Eq. (1).

This approach assumes that LOCAL mixing ratios are solely due to vehicle emissions (in reality, they may also contain point sources or other types of emissions, including those not associated with combustion). The purpose of this calculation was twofold. First, we were interested in determining whether this computationally simple approach could yield realistic fleet-average emission factors comparable to those obtained using the plume-based approach. Second, we were interested in determining EFs for pollutants not sharing a common inlet with CO2 (i.e., for benzene during the winter, and HNCO and HCN during the summer), which would allow for a seasonal comparison. Here, the assumption is that, when integrating over a sufficiently long interval of time, the majority of vehicle plumes are captured by both inlets (i.e., both the CO2 and pollutant X are detected) and that meteorology or turbulence affects the dilution of the pollutant and CO2 equally.

Figure 1 shows histograms as a function of season for the measured ambient concentrations as well as for the background (BKG) and on-road (LOCAL) contributions for (a) HNCO and (b) HCN; Fig. S4 shows similar histograms for (a) benzene and (b) black carbon. Figure 2 shows the mean BKG and LOCAL contributions to the measured ambient concentration for the four pollutants, as a function of season. For both benzene and BC, the LOCAL contribution is dominant, indicating strong traffic sources for these pollutants. We observed a small seasonal dependence in ambient benzene, with overall higher concentrations in the winter than in the summer, as observed by others (Tan et al., 2014; Lough et al., 2005). Separation of the observations into the BKG and LOCAL contributions reveals that the shift is largely in the LOCAL contribution rather than the BKG contribution, consistent with an enhanced wintertime emission factor for benzene (Tan et al., 2014; Lough et al., 2005), attributed to higher cold-start emissions and changes in fuel composition. An enhancement in wintertime benzene concentrations may also be partially attributed to an increase in benzene emissions from residential wood combustion (e.g., wood stoves, fireplaces). This enhancement would manifest in the BKG contribution (which is indeed slightly higher in the winter than the summer). On a national scale, this source is significant (CCME, 2012); however, in urban areas, wood heating is the primary home heating fuel for <0.1 % of residences (Matz et al., 2015), and so it is unlikely that this source is significant within the GTA. Increases in wintertime benzene may also be attributed to a shallower boundary layer height. A seasonal comparison is not available for black carbon.

To our knowledge, our dataset represents the first seasonal comparison of ambient HNCO measurements made at the same location. Figure 2c illustrates a decrease in HNCO concentrations from the summer to the winter. Shallower boundary layer heights would be expected to lead to enhanced wintertime concentrations if HNCO emissions and sources remain constant, and yet we observe lower concentrations of HNCO in the winter. Hence the boundary layer height is a potential issue that can reduce the apparent differences between summer and winter. Further inspection of Figs. 1a and 2c reveals that the seasonal difference is largely in the BKG contribution rather than the LOCAL contribution. The lower HNCO mixing ratios in the winter could be due to a reduction in photochemical activity and/or source strength of secondary HNCO precursors (e.g., biogenic amines) (Woodward-Massey et al., 2014; Roberts et al., 2014), or due to decreased influence from biomass burning. However, the extent to which wild fires contribute to summertime HNCO concentrations is not well established and may be less significant given HNCO's moderate lifetime (as short as a few hours in clouds, but typically weeks to hundreds of years) (Borduas et al., 2016; Barth et al., 2013; Zhao et al., 2014) and the distant location of major Canadian wildfire events relative to Toronto. Although residential wood burning could also contribute to HNCO across the GTA in the winter, a recent study by Coggon et al. (2016) showed that common residential wood fuels (e.g., heartwood and sapwood) have low nitrogen content and thus lower emissions of nitrogen-containing VOCs such as HNCO and HCN. Consistent with this finding and the low incidence of residential wood burning in the GTA (Matz et al., 2015), the HNCO BKG component is low in the winter. Rather, the LOCAL component dominates the contribution to the measured HNCO in the winter, indicating the significance of on-road emissions as an HNCO source. Although lower temperatures are thought to enhance HNCO vehicle emissions (particularly cold-start emissions) (Suarez-Bertoa and Astorga, 2016), the similarity in the magnitude of the LOCAL component between seasons suggests that, overall, the primary on-road HNCO emissions remain relatively constant.

Similar to HNCO, we observe a strong seasonal dependence for HCN. The histogram in Fig. 1b shows a much broader distribution and higher mean for the summer compared to the winter (Fig. 2d). Separation of the observations into BKG and LOCAL contributions in Fig. 1b reveals a strong seasonal difference for both components, although the difference is more striking for the BKG component. As with HNCO, despite shallower boundary layer heights in the winter, the overall concentrations are observed to be lower in the winter. The same arguments regarding the potential impact of residential wood burning on wintertime HNCO emissions apply to HCN. The large increase in BKG in the summer is consistent with the wildfire season in Canada, and that biomass burning is thought to be the major source of HCN to the atmosphere. Given the relatively long lifetime of HCN (∼2–5 months) (Li et al., 2003) compared to HNCO, biomass burning episodes in other parts of Canada would be expected to have a greater potential to influence background HCN in Toronto compared to HNCO. A strong seasonal pattern for HCN has previously been observed for tropospheric HCN column measurements (Zhao et al., 2002); seasonal measurements of HCN at an urban location have not been made. The bulk of the total measured HCN concentration is in the BKG component rather than the LOCAL component, especially in the summer, suggesting that, in relative terms, on-road HCN sources may be less significant than other regional or global sources. This is in contrast to benzene (dominant LOCAL component in both seasons) and HNCO (dominant LOCAL component only in the winter). Interestingly, examination of Fig. 1b also reveals a strong seasonal dependence in the LOCAL component, suggesting a possible seasonal dependence in the on-road HCN emissions, as discussed below (Sect. 3.3.3).

We obtained plume-based median and mean black carbon emission factors of 24.9 and 85.6 mg kgfuel-1 respectively (IQR: 10.3–76.4 mg kgfuel-1). BC emission factors obtained by prior tunnel, near-road, and mobile studies are listed in Table S9 for comparison. Literature emission factors for heavy-duty diesel vehicles range from 160–2400 mg kgfuel-1, 1 to 2 orders of magnitude higher than the literature emission factors for light-duty gasoline vehicles, which range from ∼1 to 300 mg kgfuel-1. LDGV emission factors at the high end of this range are from older studies in more polluted environments (Westerdahl et al., 2009). In many of the earlier studies, BC emission factors were obtained using an aethalometer with a 1 min time resolution, which may not have been fast enough to accurately quantify BC emission factors. In the current study, the poor performance of the time-based approach with respect to yielding BC EFs in agreement with the plume-based approach may also indicate that high-time-resolution measurements of BC and good plume definition are required to accurately estimate BC EFs from mobile measurements. However, more comparisons are needed to determine if and how calculated BC EFs depend upon the BC measurement method.

As was observed for the other pollutants (benzene, NOx, PN etc.), the BC emission factors obtained in this study are on the lower end of the reported literature range. Two studies have made recent measurements of the mixed vehicle fleet in Toronto. Wang et al. (2015) made BC EF measurements from their near-road stationary site in downtown Toronto using a photoacoustic soot photometer and report a mean EF of (35–55) mg kgfuel-1. Liggio et al. (2012) obtained BC EFs from transect driving downwind, and perpendicular to, a major Toronto highway (mean HDDV fraction ∼3.3 %). The authors report fleet-average median emission factors of 59.3 mg kgfuel-1 (IQR: 27.0–148.4 mg kgfuel-1) using a HS-LII instrument and 29.4 mg kgfuel-1 (IQR: 11.8–66.0 mg kgfuel-1) using a single-particle soot photometer. The values from these two studies (Wang et al., 2015; Liggio et al., 2012) lie between the median and mean obtained in the current study.

The lower values obtained in this study compared to Liggio et al. could be reflective of overall changes in the vehicle fleet over the past 5 years leading to reductions in BC emissions, consistent with observations at other locations (Ban-Weiss et al., 2008; Dallmann et al., 2012). The discrepancy between our study and the other two Toronto studies could also be related to location: their fixed/limited sites may not be representative of the full fleet across the GTA. Given the difference in LDGVs and HDDV BC EFs, the emission factor calculation will be quite sensitive to the frequency at which each vehicle type is sampled, which will be location dependent. This sensitivity can be quite dramatic: a recent study (Dallmann et al., 2013) found that, due to their higher associated BC emissions, even a small fraction (<1 %) of heavy-duty trucks can significantly bias the calculated LDGV emission factors (by over 40 %). For a pollutant exhibiting wide inter- and intra-vehicle variation in EFs, obtaining measurements that capture the full fleet make-up over a range of driving conditions is critical. Although we did not record the number of HDDVs (expected fraction ∼4 %), the mobile design and scope of our study helps to mitigate location-specific results. Overall, we found that the top 4 % of plumes had vehicle emissions greater than 320 mg kgfuel-1 which are typical of heavy-duty vehicles.

As previously mentioned, literature HNCO EFs have been obtained exclusively from a limited number of dynamometer studies (on both gasoline and diesel vehicles) and so a comprehensive understanding of the real-world magnitude and variability of HNCO EFs is lacking. Here we obtain the first fleet-average EFs for HNCO. Table 3 compares the HNCO emission factors available in the literature with the wintertime plume-based HNCO median EF obtained in this study (2.3 mg kgfuel-1). Our time-based analysis (Fig. 3b) suggests that HNCO EFs are similar in the summer and winter (with slightly higher EFs in the summer, contrary to the behavior of benzene).

Only two previous dynamometer studies (Brady et al., 2014; Suarez-Bertoa and Astorga, 2016) obtained HNCO emission factors from gasoline vehicles and the average EFs reported from those studies differ by more than 1 order of magnitude. HCNO was measured by Acetate-TOF-CIMS and Fourier-transform infrared spectroscopy in the former (Brady et al., 2014) and latter (Suarez-Bertoa and Astorga, 2016) studies respectively. The plume-based median EF obtained in this study is about a factor of 2 higher than that obtained by the earlier study (fleet average of 0.91±0.58 mg kgfuel-1 for eight LDGVs) (Brady et al., 2014), but significantly lower than that obtained more recently (fleet average of 93 mg kgfuel-1 for three LDGVs, or 29 mg kgfuel-1 if the anomalously high LDGVs are omitted) (Suarez-Bertoa and Astorga, 2016).

Interestingly, emission factors ranging from 0.21 to 3.96 mg kgfuel-1 were recently obtained from an engine dynamometer study on a single light-duty diesel engine, in agreement with the results from the current study (Wentzell et al., 2013). This may suggest that HNCO emissions from gasoline and diesel vehicles are of similar magnitude. In contrast, HNCO emission factors for an off-road diesel engine have been found to be 1 order of magnitude higher, and it has been suggested that the magnitude and range of HNCO emissions, as well as their dependence on operating conditions, could be different for this type of engine (larger, off-road diesel engine) (Link et al., 2016; Jathar et al., 2017). Much of the early work on HNCO vehicle emissions was prompted by the finding that selective catalytic reduction (SCR) systems could constitute an important source of HNCO (Kröcher et al., 2005; Heeb et al., 2011, 2012), but the impact of SCR systems (or other control technologies such as the diesel particulate filter, DPF, or diesel oxidation catalyst, DOC) is disputed (Jathar et al., 2017).

In addition to a wide range of emission factors, the available literature revealed conflicting information on the conditions leading to elevated HNCO emissions, as well as high inter-vehicle variability. HNCO emissions have been observed to vary by as much as 1 order of magnitude depending on the driving cycle, but the influence of hard acceleration and cold engine starting is contested (Brady et al., 2014; Suarez-Bertoa and Astorga, 2016). Similarly, studies have demonstrated opposite trends for idle vs. active operating conditions (Link et al., 2016; Wentzell et al., 2013). For all these reasons, a direct comparison of the EF obtained in this study to reported EFs is challenging. The current study cannot reveal the mechanism of HNCO production from diesel or gasoline vehicles, or its dependence on factors such as driving condition and the presence of various after-treatment technologies. However, a key strength of our study is that it is based upon a large number of vehicles operating on-road in real-world conditions, thus implicitly reflecting a range of these factors. Therefore, we suggest that the IQR reported here (1.37–4.15 mg kgfuel-1) along with the overall distribution of measured HCNO EFs (Fig. 4c) provides the most realistic constraint to date on the magnitude and variability of HNCO emissions.

As with HNCO, EFs for HCN have been obtained exclusively from a limited number of dynamometer studies. Table 4 lists HCN emission factors obtained in this study along with those obtained from prior dynamometer studies; here distance-based units (mg km-1) are used for ease of comparison. The seasonal dependence of HCN vehicle emissions has not been previously studied. Interestingly, Fig. 3c shows that the HCN EFs obtained using the time-based approach exhibit a strong seasonal dependence, with the median summertime EF almost a factor of 5 higher than the median wintertime EF. This behavior is opposite that of benzene, which has higher wintertime EFs by about a factor of 2 due to enhanced cold-start emissions and changes in gasoline composition (Lough et al., 2005). Although the mechanisms for HCN and benzene formation are different (the former involving chemistry on high-temperature catalysts), the reasons for the higher HCN EFs in the summer are not known.

Early studies on some of the first-generation three-way catalysts yielded very high HCN emission factors, typically under abnormal or malfunctioning operating conditions (Bradow and Stump, 1977; Keirns and Holt, 1978; Cadle et al., 1979; Urban and Garbe, 1979, 1980). The magnitude of the HCN emissions exhibited high car-to-car variability and a strong dependence on operating conditions, as well as the presence and composition of the catalysts. An average LDGV HCN EF of 12.1 mg km-1 was estimated from a review (Harvey et al., 1983) of these early studies – over 2 orders of magnitude greater than the EFs obtained here. However, those EF estimations were obtained for all driving modes for both normal and abnormal operating conditions.

Given the significant improvements in catalyst and emissions reduction technologies since the 1970s and 1980s, the applicability of these early studies to current HCN emission is questionable. Certainly, more recent studies (Karlsson, 2004; Baum et al., 2006; Becker et al., 1999; Moussa et al., 2016) suggest that present-day HCN EFs are much lower with individual vehicle EFs ranging from 0 to 11.7 mg km-1 (see Table 4). However, these limited dynamometer studies also reveal a large inter-vehicle variability in HCN EFs, with no clear pattern between emissions and vehicle characteristics (e.g., age). The most recent study (Moussa et al., 2016) also showed that intra-vehicle EFs are highly sensitive to fuel injection technology (e.g., gasoline direct injection, GDI, vs. port-fuel injection, PFI), after-treatment technology (presence and absence of a particulate filter), and operating conditions (e.g., aggressiveness of driving cycle, hot vs. cold starts).

The median winter HCN EF obtained in this study (using either the plume-based or time-based approach) is over 1 order of magnitude lower than the average EF obtained by the most recent dynamometer study (Moussa et al., 2016). The higher summer HCN EF obtained by the time-based analysis is in better agreement, although it is still low. However, due to the aforementioned variability in the dynamometer results, a direct comparison is not straightforward. As with HNCO, our study provides the most comprehensive HCN emission factors available to date since the mobile design allows us to obtain EFs for a large number of vehicles, thereby capturing the real-world inter- and intra-vehicle variability of emissions. Similarly, the IQR (0.32–0.88 mg kgfuel-1) and distribution of measured EFs (Fig. 4d) give new insight into the range of on-road HCN emission factors.