An Ambient Measurement Technique for Vehicle Emission Quantification and Concentration Source Apportionment Authors: Naomi J. Farren, Samuel Wilson, Yoann Bernard, Marvin D. Shaw, Kaylin Lee, Mallery Crowe, David C. Carslaw Published: 2024 ABSTRACT We develop a new technique called plume regression where fast response instruments located at the roadside are used to measure exhaust plumes of passing vehicles. The approach is used to generate highly disaggregated vehicle emissions information by vehicle type, which compares well with traditional vehicle emission remote sensing. Additionally, the technique provides valuable new information on ambient concentration source apportionment by vehicle type. The technique is flexible enough to consider a wide range of air pollutants and be deployed at roadside ambient monitoring locations. The new approach is used to quantify emissions and concentration source apportionment for ammonia (NH3) and nitrogen oxides (NOx). We find that emissions of NH3 are generally very well controlled from diesel vehicles including those with selective catalytic reduction systems that use NH3 to reduce emissions of NOx. By contrast, gasoline passenger cars are shown to be the dominant contributor to NH3 emissions, which increase with vehicle mileage. Average fuel-specific NH3 emission factors for gasoline vehicles range from 0.3 to 1.2 g kg–1, while diesel vehicle emission factors remain below 0.06 g kg–1, with the exception of Euro VI buses with the latest regulatory provisions (0.5 g kg–1). Introduction The implementation of policy measures and emission control technologies in the UK and EU-27 countries in recent decades has led to a reduction in emissions of many high priority pollutants from the transport sector, including nitrogen oxides (NOx), sulfur oxides, carbon monoxide, and particulate matter (PM).1 However, road transport emissions of pollutants such as ammonia (NH3) and nitrous oxide (N2O) have recently increased, due to their formation as unintended byproducts of aftertreatment technologies designed to minimize NOx, including three way catalysts (TWCs) and selective catalytic reduction (SCR) systems.2,3 NH3 contributes to the formation of harmful fine airborne particles in the atmosphere and excess nitrogen deposition in ecosystems,4,5 while N2O is a long-lived greenhouse gas and ozone-depleting substance.6,7 Another challenge for the transport sector is nonexhaust PM, arising from sources such as brake wear and tire abrasion. Due to increasingly widespread adoption of aftertreatment systems and increased electrification of the transport sector, these pollutants are expected to persist as the fleet modernizes. As emissions legislation becomes increasingly stringent, it is critical to develop new measurement techniques and data processing methods for accurate emissions quantification. Traditionally, remote sensing (RS) has been used as an unobtrusive technique for measuring vehicle exhaust emissions during normal, on-road operation.8,9 RS measurements have been conducted globally and the data provides real-world evidence for the efficacy of emission control technologies and vehicle emission policies, evaluating the impact of specific driving conditions, and performing fleet surveillance studies.10 The instrumentation is deployed across a single lane of traffic and individual vehicle exhaust plumes are measured over a short duration of about 0.5 s. A single vehicle RS measurement represents instantaneous emissions and carries high uncertainties as to how the vehicle performs. Therefore, multiple measurements of a single vehicle type, e.g., a Euro 5 passenger car, must be aggregated to generate an emission estimate with an associated uncertainty. Tailpipe pollutant emissions are measured as ratios to CO2, from which fuel-based emission factors (EFs) can be derived using carbon-balanced assumptions related to the combustion of the fuel.11 The main limitation of RS is the suite of pollutants that can be measured. There is limited scope to measure gaseous species of increasing concern, such as NH3, N2O and speciated hydrocarbons. RS systems lack accuracy for particulates; they are not sensitive enough to quantify PM from modern vehicles and do not measure metrics such as particle number (PN) or black carbon (BC).12 An alternative approach to RS is point sampling (PS), where continuous fast response measurements are made at the curbside and the plumes from individual vehicles are measured.13−15 By simultaneously measuring CO2 and pollutants such as NOx and integrating the concentrations, emission ratios determined. In principle, any air quality instrument with sufficient response time and accuracy can be deployed, which drastically broadens the suite of gaseous pollutants and particle metrics that can be measured compared to RS. Nevertheless there are several important practical constraints which currently affect PS. The principal constraint is having enough of a time gap between vehicles to isolate an individual vehicle plume from potential interference from other nearby vehicles. A “Goldilocks” location is preferred, that is not too busy (to avoid dispersing plumes overlapping) but busy enough to make the effort of measurement worthwhile—in practice, selecting such a location is challenging. In Knoll et al. for example, approximately 70% of the data is disregarded due to potential plume interference.11 There are also other potential issues such as the effect of meteorology, whereby a plume may not be detected at all due to an unfavorable wind direction. In this paper, we develop a new way in which PS data can be analyzed to provide vehicle-specific EFs. Rather than seek to identify and quantify individual vehicle plumes of CO2 and the measured pollutants, we consider the problem from the perspective of the likelihood of observing a plume based on an average plume profile. The approach, termed plume regression, is efficient and practical and uses data from all vehicle passes. In addition to providing highly disaggregated vehicle emissions information by vehicle type, this new approach provides valuable information on ambient concentration source apportionment. Here we deploy highly accurate, portable and fast response NH3, NOx and CO2 analyzers at the curbside and use the new plume regression technique to quantify emissions and concentration source apportionment for vehicular NH3 and NOx. Materials and Methods Instrumentation Fast response air quality instrumentation was deployed on the curbside to perform continuous measurements of vehicle plumes from passing vehicles. High resolution measurements of NH3, NOx (NO and NO2) and CO2 were made. The NH3 analyzer (model HT8700E) is a quantum-cascade laser (QCL) based system which uses wavelength modulation spectroscopy to determine in situ concentrations of NH3 at 10 Hz resolution.16 The QCL emits a continuous beam at 9.06 μm into an open-air Herriott cell with concave mirrors at each end. The open-path design avoids challenges associated with surface adsorption of NH3. Further technical details are described in Wang et al.17 and the instrument calibration procedure is described in the Supporting Information. NOx and CO2 were measured using an Iterative Cavity enhanced Differential Optical Absorption Spectrometer (ICAD) developed by Airyx.18 The ICAD uses optical absorption spectroscopy in the spectral range between 430 and 465 nm to directly measure NO2, with a t90 time of 1 s. An internal converter based on gas phase titration with O3 converts NO to NO2, allowing for total NOx and NO concentrations to be derived. Atmospheric CO2 concentrations are measured simultaneously using an in-line nondispersive infrared gas sensor. Additional technical information is described in Horbanski et al.19 A custom-built vehicle measurement device was deployed adjacent to the air quality instrumentation to record vehicle pass times. The device uses optical sensors to calculate vehicle speed and acceleration, and trigger the camera to capture an image of the rear of each passing vehicle. Automatic License Plate Recognition software (Rekor CarCheck Plus) is used to extract the license plate data from these images.20 Measurement Surveys PS surveys were conducted at three locations in the city of York, UK. Two of the measurement sites were based on the University of York campus (53.947, −1.047) and the third site was near a large retail park in an industrial area (Clifton Moor Gate, 53.987, −1.103). The surveys were carried out on weekdays between September and November 2023 during daylight hours and periods of dry weather. Ambient temperatures ranged from 1.9 to 18.7 °C, with a mean temperature of 12.9 °C. The HT8700E and ICAD were placed on a platform trolley at the curbside. The HT8700E open-air cell and the ICAD sample line (30 cm length of 1/4″ diameter PFA tubing) were positioned next to each other, close to the diluting exhaust plumes from passing traffic. The vehicle measurement device was placed approximately 1 m upstream of the analyzers and the vehicle registration plate camera was a further 5–10 m upstream. All equipment was powered using two Anker portable power stations with 512 Wh and 256 Wh capacity. Photographs of the measurement sites are shown in Figures S1–S3. The license plate data was sent to CDL Vehicle Information Services Limited to obtain individual vehicle technical information. CDL is a commercial supplier and the data is sourced from data collected as part of the UK vehicle taxation system (DVLA) and data queried from the Society of Motor Manufacturers and Traders (SMMT) Motor Vehicle Registration Information System. A vast range of information is provided, including vehicle engine size, fuel type, make and model, date of registration, and the emission standard. The mileage of a vehicle at its last annual test of vehicle safety and roadworthiness (MOT test) is provided for vehicles over three years old. A total of 11,264 vehicle passes were recorded over 9 measurement days and 95% of the vehicle registration plates were assigned vehicle technical information. The mean speed and acceleration of passing vehicles was higher at Clifton Moor Gate compared to the University campus (44.2 km h–1 and 4.8 km h–1 s–1 vs 34.5 km h–1 and 0.9 km h–1 s–1 respectively). A summary of the measured vehicle fleet composition and driving conditions is provided in Table S1. Plume Regression The approach starts with identifying when a vehicle passes the instrument(s) and considers the mean plume shape of CO2. To reduce the potential for overlapping plumes, the mean plume shape is derived from data associated with vehicle passes that have a 20 s time gap between vehicles. Considering a 20 s time gap reduces the number of measurements available (306 vehicle passes out of 10,252 in total) but provides a clear plume profile, as shown in Figure 1. The profile shown in Figure 1 varied little by the main vehicle types and pollutants (Figure S4), where there tends to be a rapid rise in concentration followed by a slower falloff; although there is less certainty with heavy good vehicles (HGVs) where there are fewer measurements. In essence, the plume profile provides a probability that a plume will be detected from a vehicle on average across all sites and conditions sampled. In some situations and with larger data sets, there could be merit in using different plume profiles e.g., for trucks with vertical exhaust pipes, which were not present in this study. Mean plume profile compiled using CO2 plumes derived from vehicle passes with at least a 20 s interval before and after other vehicle passes with 95% confidence intervals shown. Time on the x-axis is the time since the front of a vehicle is detected. The plume has been normalized so that the area under it equals 1. The mean plume profile can be used to illustrate the issue of plume overlap and interference from vehicles other than the one being measured. Taking the average plume shape in Figure 1 and adding it to the time series each time a specific type of vehicle passes generates a plot like Figure 2. A sample period of 15 min is shown where vehicles of different types pass the instruments. The choice of vehicle type grouping is only to illustrate some important issues. First, it is clear from Figure 2 that there is no single period where a vehicle plume exists in isolation and that there are always contributions from preceding vehicles. This situation is typical of PS and can only be mitigated by choosing a quieter location. Simulated plume profiles for a typical 15 min PS measurement period, created using the average plume profile shown in Figure 1. The colors of the stacked plume profiles represent different vehicle types, chosen to illustrate both the importance of plume overlap and the analysis approach that is developed. The vertical lines on the x-axis show the times when a vehicle passes based on when the front of the vehicle blocks the light beam. Second, Figure 2 shows that at any point in time there can be a range of contributions from different vehicles depending on both the type and proximity of a vehicle to the one being measured. The analysis approach adopted is as follows. The vehicle categories are decided ahead of the analysis e.g., by fuel type and broad vehicle category, such as those shown in Figure 2. For each category a column is added to the time series data set, which is initially set to zero. Each time one of the vehicles in the category passes the instrument, the normalized plume profile is added to the column for that particular vehicle category. This process is repeated for all vehicles that pass the instrument. Using Figure 2 as an example, 11 vehicle categories are considered. The sum of any category column is equivalent to the total number of vehicles in that category, as the area under each plume equals one. The second step is to calculate an increment in concentration above local background for each pollutant, which represents the contribution from vehicles. There are various ways this can be approached, but a simple rolling average low percentile was chosen21,22 with a time width of 100 s. A width of 100 s was selected to comfortably straddle the typical width of a plume (Figure 1) but avoid capturing contributions other than a passing vehicle. The next step is to relate the 1 Hz time series of pollutant concentration increments to the plume weighting categories from the first step to calculate the contributions from different vehicle categories that best explain the increment in concentration of the pollutant in question. There are several options for regression, but robust linear regression has been chosen to avoid the potentially influential effect of outliers(MASS).23 While ordinary linear regression provides mostly similar results to robust regression, outliers can invalidate the assumptions used in linear regression e.g., normally distributed residuals. There is close to zero correlation between categories of vehicles, which is beneficial from a regression perspective where multicollinearity can be problematic. Low correlation is expected because vehicles are separated in time (even if there is plume overlap) and there is effectively a random order of plume overlap by vehicle type e.g., a Euro 5 gasoline car does not always follow a Euro 5 diesel car. The inherently low correlation between vehicle groupings also means that the regression models can have many terms (tens or more) and still distinguish between them with low standard error. The regression equations take the form:123where, , and are the concentration increments for CO2, NOx and NH3, respectively. The P terms represent the plume weightings for each class of vehicle considered. The a, b, and c terms represent the regression coefficients to be determined and represent the amount a normalized vehicle plume has to be multiplied by to give the best overall fit to the data for each vehicle category. The regression coefficients are then used to calculate other quantities such as fuel-based EFs derived from the molar ratio of NOx or NH3 to CO2 and assumptions concerning the carbon content of fuels. The approach above has significant benefits over approaches that attempt to isolate individual vehicle plumes. First, all the data and every vehicle pass is used in the regression approach, whereas typically only 10–30% of the data is used in the former approach owing to plume overlap. This aspect of the regression plume approach also greatly simplifies the analysis, as no time needs to be spent trying to identify and isolate individual vehicle plumes. The regression approach accepts that there is plume overlap but uses that information directly to determine relative contributions to concentrations by vehicle type. There is an interesting comparison that can be made with traditional RS where typically a measurement of duration 0.5 s is made of individual exhaust plumes. The RS approach has the benefit of being vehicle-specific and providing a measurement of the plume immediately after exiting from the exhaust. However, RS still provides a very short duration snapshot of vehicle exhaust concentrations and is therefore an inherently noisy measurement on an individual vehicle basis, regardless of the accuracy of the instruments used. In contrast, the plume regression approach has a much longer sampling duration of ≈20 s based on the plume profile shown in Figure 1. However, the “noise” is related more to plume interference. We return to this aspect of plume regression and RS in the section Comparison with Remote Sensing where the emission uncertainties of each approach are considered. Similar to traditional remote sensing, a single measurement of a vehicle using the plume regression approach is insufficient to characterize its emissions. Data Processing To account for the different response times of the HT8700E and the ICAD, the 10 Hz NH3 signal was convoluted using the base R convolve function, prior to the regression analysis. This approach generated NH3 plume profiles with a similar shape to the NOx and CO2 plume profiles. Further details can be found in the Supporting Information. Robust linear regression was performed in R using the MASS package.23 The NH3, NOx and CO2 regression coefficients for each vehicle category were used to determine average molar NH3/CO2 and NOx/CO2 emission ratios. Fuel-specific NH3 and NOx EFs, expressed in grams of pollutant per kilogram of fuel consumed, were derived from the molar ratios by assuming CO2 EFs of 3.16 and 3.17 kg CO2 per kg of fuel for diesel and gasoline vehicles, respectively.24 For the concentration source apportionment analysis, per vehicle contributions of each vehicle group to roadside NO2 increments were determined by multiplying the regression coefficients by the plume weightings for each class of vehicle considered. Results and Discussion Road Vehicle Emission Factors The PS instrumentation was deployed at three locations in York (UK) and over 11,000 vehicle passes were recorded. The data was processed using the plume regression approach and the regression coefficients were used to generate the NH3 and NOx EFs shown in Table 1. The regression model outputs are provided in Tables S2–S4. Unlike previous PS studies, none of the data was discarded, which maximized the sample sizes (n) shown in Table 1. It should be noted that distance-based emission factors (such as g km–1) can be estimated from fuel-based emission factors, which enables comparison against vehicle emission limits as described in Davison et al. (2020).25 Table 1Mean NH3 and NOx Fuel-specific EFs Grouped by Vehicle Type, Fuel Type and Euro ClassaVehicleFuel typeEuro classNH3 (g kg–1)NOx (g kg–1)nbusdieselVI0.51 ± 0.020.56 ± 0.11118cardiesel<40.02 ± 0.0312.9 ± 1.45112cardiesel40.04 ± 0.0311.5 ± 1.15312cardiesel50.03 ± 0.0110.6 ± 0.47934cardiesel60.04 ± 0.017.83 ± 0.431137cardiesel6*0.01 ± 0.024.01 ± 0.51214cargasoline<40.58 ± 0.073.40 ± 0.54288cargasoline40.90 ± 0.093.72 ± 0.54510cargasoline50.55 ± 0.042.03 ± 0.281048cargasoline60.52 ± 0.021.35 ± 0.161578cargasoline6*0.28 ± 0.021.35 ± 0.21836hybrid cargasoline60.33 ± 0.051.47 ± 0.40741HGVdiesel 1) while elevated NH3 emissions are associated with fuel rich conditions (λ < 1).2,37 The Handbook of Emission Factors for Road Transport (HBEFA) provides EFs for all current vehicle categories for a variety of traffic situations and is used widely across Europe for national inventory reports and air quality modeling.38 The latest version (HBEFA 4.2) contains emission deterioration functions for CO, NOx and NO2 but a complete update of mileage correction factors is expected to be included in HBEFA 5.1 (due for publication in mid-2025). It is important that NH3 deterioration functions are closely scrutinized prior to this, especially considering the shift to gasoline since the Dieselgate scandal (6.5% of cars registered for the first time in the UK in 2023 were diesel compared to 48.3% in 2015).34 Third, Figure 4 shows that NOx emissions from diesel vehicles generally decrease with increasing Euro class and NH3 emissions remain low, but Euro VI buses are an exception to this trend. Modern diesel vehicles are equipped with SCR systems, which use an urea-based solution to reduce NOx. However, the injection of urea must be carefully managed to avoid “ammonia slip”, whereby unreacted NH3 from the urea is emitted from the tailpipe. In this study, the majority of SCR-equipped diesel vehicles have negligible NH3 emissions even when NOx emissions are low (such as Euro 6 cars and LGVs) but the Euro VI buses have a high NH3/CO2 ratio of 0.42 ppb/ppm. Based on the vehicle manufacturer dates, a high proportion of the measured buses are expected to be Euro VI stage E buses. Compared to Euro VI stage D, these have improved regulatory provisions aimed at reducing NOx emissions during cold start situations and at lower engine load.35 However, this involves injecting the urea-based solution into the SCR at lower temperatures, which may lead to more frequent occurrences of NH3 slip, especially under urban driving conditions, which have a higher proportion of stop-start, low speed and engine idling events. Concentration Source Apportionment Aside from the quantification of road vehicle emission factors, an important benefit of PS over RS is that it provides a route to concentration source apportionment. As PS measurements produce a high resolution time series of measured pollutant concentrations at the roadside, the increment above local background, also known as the roadside increment, can be determined for each pollutant. The roadside increment essentially provides an estimate of the concentration of a pollutant that can be attributed to vehicles in the local vicinity, i.e., the traffic driving directly past the PS instrumentation. An additional step made possible by the plume regression approach is to apportion the roadside increment to particular vehicle types, using the coefficients (represented by a, b and c in eqs 1–3) output by the regression model. Here we consider source apportionment of the roadside increment of NO2 (measured directly by the ICAD at the roadside). Due to the harmful effects of NO2 on human health, many countries (including the UK) set air quality objectives for NO2 to aim to reduce the amount to which we are exposed.39 The median roadside increment of NO2 for the PS measurements was 1.4 ppbv, or 2.7 μg m–3. Figure 5 shows the average roadside NO2 increments on a per vehicle basis for different vehicle categories. The vehicle categories are chosen when the model is constructed and the NO2 increments for each category are generated using concentration data that is collected only when these vehicles pass the PS instrumentation (≈25 s of data for each vehicle pass, as demonstrated in Figure 1). Mean per-vehicle roadside NO2 increment concentrations for different vehicle groups, determined using the plume regression approach. The percentages show the contribution of each vehicle type to the total roadside NO2 increment. * refers to vehicles which have undertaken RDE testing. Figure 5 shows that on a per vehicle basis, pre Euro VI diesel HGVs contribute to the highest amount of NO2 exposure at the roadside. Other significant contributors include Euro 4 diesel LGVs and pre Euro 4 diesel cars. There is a clear distinction between the contribution of diesel and gasoline vehicles; many gasoline vehicles such Euro 5 and 6 cars contribute significantly less to roadside NO2 exposure on a per vehicle basis. An important consideration is vehicle fleet composition. The per-vehicle NO2 increments were scaled by the fleet composition to calculate the percentage contribution of each vehicle type to the total roadside NO2 increment, as shown by the percentages next to each data point in Figure 5. The percentages will vary according to the measured vehicle fleet composition at a given location. Although pre Euro VI diesel HGVs contributed most to NO2 increments on a per vehicle basis, they made up just 0.4% of the fleet in this study and therefore had a low contribution to the total vehicular NO2 increment (2.1%). Euro 5 and Euro 6 diesel cars comprised 9.2% and 11.2% of the measured fleet respectively, and were found to be the dominant contributors to the total NO2 increment (23.9% for Euro 5 diesel cars and 17.9% for Euro 6 diesel cars). The percentage contribution to roadside NO2 increment for vehicle types such as Euro 6 gasoline hybrids and Euro 6 RDE gasoline cars remained low (≈1–2%) despite their significant contribution to the fleet composition (7.3% and 8.3% respectively). Recommendations and Outlook Looking ahead, the plume regression approach could be widely adopted around the world, through the deployment of fast response analyzers measuring a wide range of air pollutants and CO2 at roadside monitoring sites. This approach offers a potentially cost-effective solution for cities to upgrade their air quality monitoring stations to better understand transport emissions and develop data-driven policies. The existing capability of roadside monitoring sites to provide hourly air pollution concentrations for comparison with air quality health based thresholds would significantly be enhanced by also providing vehicle emissions quantification and determining how much different vehicles are contributing to ambient air pollution concentrations. It is recommended that a program of renewing roadside air pollution monitoring considers deploying fast-response analyzers e.g., ICAD but paired with an ANPR camera for vehicle fleet apportionment. Measurements could be conducted in a pollution hotspot or area proposed to be a low emission zone, and roadside pollutant increments scaled by the location-specific vehicle fleet composition to establish which vehicle types to minimize or exclude to have the most beneficial effect on ambient exposure to harmful pollutants. Direct correlation between fuel-specific EFs and exposure concentrations for given vehicle types is not anticipated, due to complexities associated with vehicle-induced turbulence and plume dilution. Therefore, it is critical that new techniques, such as the point sampling and plume regression approach presented in this study, are developed in order to apportion emissions and exposure concentrations of a broad suite of pollutants to a wide range of vehicle types.