Classified traffic volume and speed study of Delhi (Malik et al., 2018) provides traffic count and speed for the roads of Delhi based on the traffic volume and speed measurements conducted at 72 locations (Fig. 1) over Delhi in the year 2018 as a part of TRIPP of IIT Delhi. We will refer to this dataset as TRIPP data from now on. TRIPP data provide hourly traffic from 08:00–14:00 IST (Indian standard time) for eight fleet types (2W, 3W, cars, buses, minibuses, HCV, LCV and non-motorized vehicle; NMV) on more than 12 000 major road links over Delhi (Malik et al., 2018). These road links are further classified into five road classes (RClass1 to RClass5) based on the width of the road (Table S2). More detail of TRIPP traffic flow and its methodology is available elsewhere (Malik et al., 2018, 2021). As the TRIPP data are only available for 08:00–14:00 h, we use the speed–flow–density relationship by Malik et al. (2021) to estimate the hourly traffic for each road link in Delhi.

The six types of primary vehicle categories (2W, 3W, CAR, BUS, LCV and HCV) have been further classified into 127 categories (Table S1) according to fuel, engine capacity and emission standards to match the COPERT-5 vehicular classification. The fuel share of petrol/gasoline, diesel and CNG (compressed natural gas)/LPG vehicles in Delhi for passenger and freight vehicles has been obtained from Dhyani and Sharma (2017) and Malik et al. (2019), respectively. The engine share for primary vehicle categories has been taken from working papers (Sharpe and Sathiamoorthy, 2019; Anup and Yang, 2020; Deo and Yang, 2020) of the International Council on Clean Transportation (ICCT). In India, the emission norms/standards, known as Bharat Stage (BS) which can be considered equivalent to the European emission standards – Euro, have been introduced in a phased manner. These norms were introduced for passenger cars and later extended to other vehicle categories. For example, the BS-I (India-2000) for passenger cars was implemented in 2000 followed by BS-II, BS-III and BS-IV in 2005, 2010 and 2017, respectively. The BS-VI for passenger cars has been recently introduced in 2020; therefore, it has not been considered in our study. For Delhi, the timeline of BS implementation for passenger cars and other vehicles are shown in Table S3. The vehicles prior to the implementation of BS norms have been considered as conventional (or BS-0 for simplicity). The BS share of the vehicles has been derived using the survival function method described in (Goel et al., 2015; Malik et al., 2019). The vehicle survival was calculated for the past 20 years by considering 2018 as the base year and then the BS share was calculated based on the age of the vehicle with respect to 2018 (Table S4). The final share of the primary vehicle category as per fuel, engine and BS norms has been calculated by multiplying the fuel share, engine share and BS norms share and shown in Table S1. In this study, BS and EURO/Euro have been used interchangeably, and BS-I to BS-IV, BS1 to BS4 or EURO1 to EURO4 represent the same emission standard.

The emission factor (EF) is a crucial parameter needed for emission estimation. Road traffic vehicular emission depends on a variety of factors such as vehicle type, fuel used, engine types, driving pattern, road type, emission legislation type (BS/EURO) and speed of the vehicle. We have adopted the recent COPERT-5 Tier-3 methodology and used the speed-based emission factor (https://www.emisia.com/utilities/copert/, EMISIA, 2021, last access: 10 October 2021) for 127 vehicle types (Table S1) and according to the emission legislation up to BS/EURO-4 (since BS-VI is not implemented in 2018). The EF as a function of vehicle speed (v) is calculated using Eq. (4): 4EF(v)=α×v2+β×v+γ+δvε×v2+ζ×v+η, where v is the speed and α, β, γ, δ, ε, ζ, and η are coefficients that vary with vehicle type.

The coefficients for each pollutant and vehicle category are taken from the COPERT-5 database (COPERT-5 guidebook, 2020). The emission factors are further corrected for the emission degradation occurring in older vehicles considering the mileage as discussed in the COPERT-5 guidebook (2020). COPERT relies on mean driving speed and travel distance. The mean speeds are relatively low under urban driving conditions, and emission factors are highly variable within this speed range due to the speed fluctuations caused due to real-time driving behaviour (frequent braking, acceleration, deceleration, idling). Lejri et al. (2018) have estimated the relative errors on fuel consumption and NOx emissions related to mean speed variations from 2 to 10 km h-1 and estimated errors up to 25 %–30 % in fuel consumption and NOx emissions. Therefore, to account for the emissions due to the speed fluctuations around the mean speed, a factor of 1.2, i.e. 20 % increase has been applied to the final dataset. This has been applied for all the hours and all the pollutants. Although we apply the same factor for all hours of the day, the added emissions are more during high congestion hours and less during low congestion hours.

The non-exhaust emissions (Singh et al., 2020) have not been calculated in this study. As COPERT does not provide the EFs for the 3W CNG category, we have used EFs of CNG mini car for this. The BC and OM emissions are computed using the fraction (by COPERT-5 guidebook, 2020) from the PM exhaust. We have compared the COPERT EFs used in this study with the earlier reported EFs and shown in Table S12 to elaborate upon the potential uncertainty in the key vehicle categories. Further, the emission uncertainties have been discussed in Sect. 4.

The model calculates hourly emissions for each road link of finite length and uses hourly traffic volume and emission factors as a function of speed for 127 vehicle categories (Table S1). The hourly emission rate (Q) for each road link is calculated using Eq. (5). The total emission for a given hour is calculated by taking the sum of emissions across all vehicle categories: 5Qi,hp=∑jVi,j,h×EFjpvi,h×Li, where Qi,hp is emission rate of a pollutant p for road link i and at hour h, where h=0 to 23; Vi,j,h is the traffic volume of vehicle category j for road link i at hour h, where j=1 to 127; Li is the length of road link i; and EFjpvi,h is the emission factor of pollutant p for vehicle category j as a function of speed vi,h for road link i at hour h.

The hourly emissions have been calculated for each pollutant over each road link then gridded at 100 m × 100 m resolution using the methodology described in Singh et al. (2018, 2020) to produce the hourly gridded emission inventory for Delhi.

The estimated hourly traffic volume (in PCU) and speed profiles for Delhi are shown in Fig. 2. An anticorrelated diurnal variation is seen in the traffic volume and speed. The traffic volume during weekdays tends to have a bimodal profile with a morning peak (09:00–11:00 IST) and an evening peak (18:00–20:00 IST). A similar traffic volume profile has also been observed by other studies over Delhi (Dhyani and Sharma, 2017; Sharma et al., 2019). Similar bimodal traffic profiles are also observed over the cities around the world subject to the city specific travel demand (Järvi et al., 2008 for Helsinki; Jing et al., 2016 for Beijing). The evening peak traffic volume tends to be 40 % higher than the morning peak. The vehicular composition changes hourly (Fig. S1) and also varies with respect to the road classes (Table S5). The nighttime goods vehicle share is more in comparison to the passenger and personal vehicles (Fig. S1). The weekend traffic volume does not show a morning peak due to closure of the offices/workplaces and shows evening peaks due to shopping and other weekend activities. As usual, the minimum traffic volume is observed at night (00:00–04:00 IST) because of the reduced human and commercial activities. Due to the minimum traffic at night, the traffic moves with an average speed of 51 ± 6 km h-1 with almost no congestion. As traffic volume increases, it starts to build congestion, leading to reduced speed. The average speed during morning peak hours of the weekdays is estimated to be 30 ± 14 km h-1, whereas the evening speed is estimated to be 28 ± 15 km h-1. The evening congestion leads to an average 46 % reduction in the average speed, increasing the travel time by a factor of 2. We calculated the average profiles for each road link by combining weekdays and weekends and used them in the emission calculations. The estimated profiles averaged across all road links are shown in Fig. 2. We have estimated 27, 31, 6, 1.7, 0.95 and 3.14 billion VKT driven by CAR, 2W, 3W, BUS, HCV and LCV, respectively. The comparison between estimated annual VKT and reported by other studies is tabulated in Table S11. This comparison table includes the studies which have either reported annual VKT or have provided enough data to calculate annual VKT. The VKT values compare well with the earlier studies by considering the fact that the uncertainties exist in the method of estimation, year and study domain. Malik et al. (2019) estimated the destined and non-destined VKT of freight vehicles (HCV and LCV) with the actual measured traffic at several entry points in Delhi. Goel et al. (2015) estimated the annual VKT based on the annual mileage of the 2W and cars obtained from PUC (pollution under control) certification data and the number of registered vehicles. The VKT reported by Goel et al. (2015) for CARS and 2W are slightly lower than our study. The study by Goel et al. was conducted in 2012. Since then, the car and taxi share has almost doubled in Delhi due to increased travel demand and economic growth (DDA, 2021). The study by Kumar et al. (2011), which is for 2010, reported higher VKT for buses and HCV as compared to the one estimated by the current study. Their estimates were based on the assumed distance travelled by each vehicle and the number of registered vehicles than the actual on-road vehicle. Guttikunda and Calori (2013) reported high VKT for buses and HCV. The study by Sahu et al. (2011) for NCR Delhi estimated very high VKT for 2W and cars. While earlier studies have reported different VKT values, the relative VKT share compares well with our study. Moreover, the VKT estimated by recent studies are close to our estimates.

A multi-pollutant hourly and high spatial resolution (100 m × 100 m) emission inventory has been prepared for Delhi. As an example, the spatial distribution of NOx emission at 03:00–04:00, 09:00–10:00, 15:00–16:00 and 18:00–19:00 IST, representing early morning, morning peak, afternoon and evening peak, respectively, is shown in Fig. 3. The emission rate during the evening peak hours is the highest during the day, followed by morning peak hours. The high traffic volume along with traffic congestions lead to more emissions during the peak traffic hours (Jing et al., 2016). The emission during the afternoon hours is comparable or less than that of the morning hours, whereas the early morning emissions are lowest because of low traffic volume moving with free flow speed. The diurnal profile of emissions has been discussed in detail in Sect. 3.5.

The annual emissions have been calculated by summing the hourly emissions to get daily emissions and then multiplying it with 365 (number of days in a year) to get annual emissions. The monthly variation in the emission has not been considered as the monthly variations are much smaller than the hourly variations. We estimated an annual emission of 1.82 Gg for PM, 0.94 Gg for BC, 0.75 Gg for OM, 221 Gg for CO, 56 Gg for NOx, 64 Gg for VOC, 0.28 Gg for NH3, 0.26 Gg for N2O and 11.38 Gg for CH4 in 2018.

The hourly emissions over Delhi have been summed together to calculate the daily emissions for all the pollutants. The spatial variation of daily mean emission rate has been analysed over three selected regions, viz. inner, outer and east side Delhi  (as shown in Fig. 1). The total emission for each pollutant and for each region has been tabulated in Table S6. The outer Delhi region has the highest emission (51 %–53 %) for all the pollutants because of its largest area of 1106 km2 which is 4.5 times of inner Delhi. To avoid the influence of area on the emissions, we have calculated the emission flux (i.e. emission per unit area) and shown in Table S7. The emissions flux is highest for inner Delhi, followed by east side and the outer Delhi region. For all pollutants, the emissions flux in inner Delhi is 40 %–50 % higher than the average emission of Delhi, whereas the emission flux in outer Delhi is ∼ 46 % lower. The emission flux is consistently high along the grids containing major roads (Fig. 3), intersections and major business hubs. Inner Delhi consists of major business hubs, workplaces and government offices, which entertain more vehicular activity in this region resulting in congestion that leads to reduced speed and enhanced emissions. The daytime average speed across all roads in inner Delhi is 29 km h-1 which is lower than the daytime average speed of 32 km h-1 in outer Delhi. The lower speed and higher traffic density influences the economic driving behaviour, resulting in frequent braking, idling, acceleration and deceleration that enhances the vehicular emission. Moreover, the morning and evening peak hours with higher traffic and lower speed have the highest emission as compared to the rest of the day. In these heavy congested hours, the vehicle is forced to run in lower speed which boosts the emission.

The emissions along the five road classes used in this study have been calculated and shown in Table 1, and the hourly variation of emission has been shown in Fig. 4. The RClass3 has a substantial emission share (∼ 35 %) across all pollutants followed by RClass5 and RClass2, whereas RClass1 holds the minimum emission share (∼ 2 %–3 %). The dominant emission share of RClass3 is due to the optimum vehicular activities over the longer road length. RClass2, which is the class of feeder roads to RClass3, RClass4 and RClass5, contributes ∼ 23 % to the emission. The multi-lane wider roads, RClass4 and RClass5, contribute ∼ 13 %–15 % and ∼ 21 %–25 % respectively to the total emission. To remove the dependency of the road length, we calculated the emission per kilometre segment of a road. The emissions (per km) over multi-lane wider roads (RClass4 and RClass5) are almost 2 times of the RClass3 (Table S8 and Fig. S2) due to more traffic flow, irrespective of the congested conditions. However, the emission per lane per kilometre (Table S9) for RClass1 is found to be the highest because of lower speed and congestion and major share of 2W. This shows that effective management of traffic in narrow roads to reduce the congestion will be beneficial in reducing the pollution without impacting the traffic volume. The multi-lane wider roads (RClass4 and RClass5) help the vehicle to maintain an economic speed resulting in minimum congestion and lower emission; however, they are the emission hotspots in Delhi.

Dynamic traffic volume and speed, as discussed in Sect. 3.1, results in diurnal variation in the emissions during a day. Figure 4 shows the hourly emissions (Mg h-1) and contribution of each road class at each hour in Delhi. The temporal evolution of emission is linear with the traffic variation in a day with the minimum variation during the night-time and remarkable variation during the human active hours (08:00–20:00 IST). Among different road types and for all the pollutants, RClass1 has the lowest and RClass3 has the highest emission proportional to the traffic volume. A similar temporal variation of the NOx emission rate is observed in a study, for different road types of Beijing (Jing et al., 2016). For most of the pollutants (except PM, BC and NOx), daytime (08:00 to 20:00 IST) contributes ∼ 70 % to the daily emissions whereas the morning (09:00 to 11:00) and evening (18:00 to 20:00) rush hours alone altogether add 30 %–40 % to the total emissions. The increasing activity of goods vehicles (HCV + LCV) during afternoon and nighttime (Fig. S1) elevates the emission of PM, BC and NOx from these vehicles (Fig. 5), resulting in a different diurnal profile compared to other pollutants. The emissions of NOx and particulate pollutants (PM and BC) during late night hours (11:00–05:00 IST) is relatively higher, adding up to 60 % and 75 % of total particulate and NOx nighttime emissions, respectively, as shown in Fig. 5. The contribution of vehicle type has been discussed in detail in Sect. 3.6. The diurnal evolution of emission is also visible in the hourly spatial map shown in Fig. 3. Early morning with minimum traffic volume has lower emission whereas the evening rush hour with increasing congestion has higher emission. The density of higher emission grids (Fig. 3) in the inner Delhi region is higher compared to other regions throughout the day.

The percentage share of major vehicle types to the total emission of nine pollutants has been calculated and shown in Table 2; and its hourly contribution is shown in Fig. 5. The 2W vehicles, having a major vehicular share (Table S5), are the major contributors to the total emissions for all the pollutants, except for BC, NOx and N2O. The goods vehicles (HCV and LCV) contribute substantially, mainly during nighttime, to the PM, BC and NOx emissions. Buses have the highest contribution to NOx emissions and substantial contribution to PM, BC and CH4. Cars are the dominant source for NH3 and N2O and contribute substantially to PM, BC and NOx emissions. However, most of the emissions are from diesel cars.

The vehicular fuel share to the total emission for each pollutant is shown in Table 3. Petrol vehicles are the largest contributors to the CO (∼ 94 %), VOC (91 %), NH3 (86 %), OM (74 %), CH4 (67 %) and N2O (58 %), whereas diesel vehicles are the largest contributor to the BC (∼ 80 %), NOx (59 %) and PM (54 %) emissions. The contribution of the CNG vehicles is relatively smaller, except for the NOx and CH4, where they contribute to ∼ 30 %, almost one third, of the total emissions. The larger contribution of petrol to the VOC, CO, OM and CH4 emissions are dominated by 2W where we estimated that 2W in Delhi alone contribute 90 %, 88 %, 71 %, and 66 %, respectively, as shown in Table 2. The contribution of 2W is also highest to PM (42 %). The larger share of 2W towards the CO emissions has also been reported earlier, 61 % in Goyal et al. (2013); 43 % in Sharma et al. (2016) and 37 % in Singh et al. (2018). Higher emission share of 2W is due to the higher emission factor of VOC in petrol-fuelled 2W (Hakkim et al., 2021) that has also been reported in a multi-year emission study over Delhi by Goel and Guttikunda (2015). The PM emissions are dominated by diesel-fuelled HCVs (16 %), LCVs (13 %), buses (14 %) and cars (∼ 13 %), whereas 2W are the main source in petrol-fuelled vehicles contributing ∼ 42 % to the total PM emissions. Earlier, Sharma et al. (2016) reported 33 % share of 2W emissions in 2014. The share of petrol cars and CNG buses towards the PM, BC and OM emissions is less than 2 %. While it is clear that diesel-powered vehicles are the major source of PM emission, earlier studies have reported similar results but with large variations of HCVs in emission share. The largest share of diesel-fuelled HCV is reported as 92 % by Goyal et al. (2013), 46 % by Sharma et al. (2016) and 33 % by Singh et al. (2018). All these studies reported minimal emission share (less than 10 % combining both diesel and petrol cars). The largest share of HCV, LCV and diesel cars to BC emission is because of higher emission factors (Zavala et al., 2017) contributing to total urban BC emission as shown by Bond et al. (2013). The petrol cars contribute more than half of the total NH3 emissions and among them, the Euro 2 with higher emission factor has the largest share of 39 %. The diesel vehicles (HCVs, LCVs, diesel buses and cars) altogether contribute significantly to the PM, BC and NOx emissions. The higher emission factor of diesel-fuelled vehicles (Wu et al., 2012) clearly reflects in the emission share. The CNG buses have the highest share (27 %) in NOx emission and around 23 % in CH4 emissions. The highest share of CNG is due to the higher NOx emission factor for CNG vehicles compared to petrol vehicles (Dimaratos et al., 2019). The larger share of ∼ 15 % from CNG buses to the total traffic NOx emission is also reported in a study of CPCB (2011). In terms of Euro or BS standard, Euro 3 vehicles have the highest share (Table S10) in the total emission, except for N2O and NH3. This is mainly because of the highest share of Euro 3 vehicles in 2W, buses, HCV and LCV (Table S4 in the Supplement). In the case of N2O, the emissions are dominated by Euro 4 cars which have around 84 % share to the total cars. For CH4, the highest share of Euro 3 vehicles is due to the higher emissions from Euro 3 2W as the emission factor of petrol vehicles is higher (Clairotte et al., 2020).

In order to have a clear picture of the dominant polluting vehicle categories, we grouped different vehicle types into 35 categories and calculated the percentage share to the total emission of 9 pollutants as shown in Table 4. We further identified the top five polluting vehicle categories for each pollutant and tabulated it in Table 5. For PM, the top five polluting vehicles account for 55 % of the total emissions, which is dominated by petrol Euro 3, petrol 2W and Euro 3 diesel HCVs. The BC emission is mainly driven by Euro 3 diesel HCVs, LCVs, buses and the top five polluting vehicles account for 66 % of the total emissions. The OM, CO, VOC emissions are dominated by 2W and the top five accounts for 71 %, 89 % and 91 % of total emissions, respectively. Petrol-fuelled cars and 2W hold the dominant share of NH3 emissions because of the larger EF compared to other categories (COPERT-5 guidebook, 2020). For N2O, 2W Euro 3 holds the highest share of 21 %, followed by EURO 4 diesel and petrol cars. The top five contributors to CH4 emissions account for 86 % of the total emissions, which are dominated by 2W and CNG buses. These two categories of vehicles altogether contribute to ∼ 97 % of the emissions.