The emission inventory for each pollutant (CO, VOCs, NO_x, PM, NH₃) for passenger cars was estimated with the following equation: 1Ej=∑i,p,k,tVp,i,k,t×BEFp,j×φp,i,j,t×γp,i,t×λp,j×θp,j×10-92Vp,i,k,t=VPp,i,k,t×VKTp,i,t3TPEs=∑jEjPEVj In Eqs. (1) and (2), E represents pollutant emissions (t); i represents the grid; j represents the pollutant type, including CO, PM, VOCs, NO_x and NH₃; k represents the vehicle type, including small passenger cars and mini passenger cars; p represents the province; VP, VKT, and BEF represent the vehicle population, annual average vehicle kilometers traveled and baseline emission factors, with units of vehicles, km yr⁻¹, and g km⁻¹, respectively. φ is the environmental correction factor (including temperature correction factor, humidity correction factor, and altitude correction factor); γ is the speed correction factor; λ is the deterioration correction factor; θ represents other correction factors (including sulfur content correction factor and ethanol blending correction factor). In Eq. (3), TPEs represents the total pollutant equivalents (kt) (For assessing the extent of environmental and techno-economic hazards posed by various pollutants), and PEV represents the pollutant equivalent value (kg) (MEE, 2018). This study focused on gasoline-fueled passenger vehicles. The vehicle kilometers traveled (VKT) of light-duty gasoline passenger vehicles (LDPVs) in 2019 were obtained from Ma et al. (2022). Based on the China Statistical Yearbook 2004–2019, we calculated the proportion of passenger vehicles by emission standard, the total number of passenger vehicles in each province in 2019 (Fig. S1 in the Supplement), and the regional distribution of small and mini passenger vehicles by emission standard (Fig. S2). The baseline emission factors were obtained based on other literature (Wen et al., 2023; MEE, 2014; EEA, 2019), with specific details provided in Table S1 in the Supplement. Furthermore, the definition of urban and rural areas involved in this study was based on the research of Li et al. (2020b).

The core difference between the traditional algorithm and the improved algorithm in this study was that the former assigned a fixed speed correction factor for each of the five speed intervals with monthly-scale activity levels, while the latter obtained continuous speed correction factors and further refined the activity levels by combining the congestion index with the traffic flow model. The detailed correction values could be found in Table S2.

For time allocation for vehicle activity levels, the following formula can be used: 4Vp,t,k=Cp,t×Vp,k=Cp,t×VPp,k×VKT where Vp,t,k is the vehicle activity level of model k in province p in time period t (vkm); Vp,k is the number of vehicles of model k in province p (vkm).

Cp,t is the time allocation coefficient of province p at time period t. It is obtained by combining the congestion delay index obtained from Baidu Map Traffic and Travel Big Data Platform (https://jiaotong.baidu.com/, last access: 27 April 2026) with the three-parameter model of traffic flow (Wang et al., 2024), which is constructed as follows:

The congestion delay index (λ) is defined as the ratio of the actual time spent by residents on one trip to the time spent in a smooth state when the travelling distance is the same, and its calculation formula is: 5λ=TTf=LvLvf=vfv Where, λ is the congestion delay index, dimensionless; T is the actual time spent travelling (h); Tf is the time spent travelling at the smooth speed (h); L is the length of the road section (km), v is the actual travelling speed (km h⁻¹), and vf is the smooth speed of the vehicle (km h⁻¹), then: 6v=vfλ Combined with the basic three-parameter model of traffic flow, the relationship between flow rate and congestion index can be derived: 7Q=Kjv-v2vf=Kjvfλ-vfλ2vf=Kjvfλ-1λ2 Where Q is the flow rate (veh h⁻¹); λ is the congestion delay index; Kj is the congestion density (veh km⁻¹); vf is the unimpeded vehicle speed, and Kj and vf are constants.

In turn, we obtain the formula for calculating the time allocation coefficient of motor vehicle emissions: 8Ct=Qt∑1tQt=Kjvfλt-1λt2∑1tKjvfλt-1λt2=λt-1λt2∑1tλt-1λt2 Where Ct is the motor vehicle emission time allocation coefficient, dimensionless; t is time; and Qt denotes the volume of traffic in time period t.

Regarding the spatial allocation of the vehicle activity level, this study was done using the road length data provided by Golder Maps as an allocation index (Gómez et al., 2018), and the gridded vehicle activity level at 0.01° × 0.01° resolution was calculated according to the following formula: 9Vp,i,k,t=Lp,i∑iLp,i×Vp,t,k where Vp,i,k,t is the activity level of model k in the i grid of province p in time period t; Vp,t,k is the number of vehicles of model k in province p in time period t; and Lp,i is the length of the road in the i grid of province p (km).

To verify the superiority of the improved inventory, this study employed the WRF-Chem model to simulate the atmospheric concentrations of PM_2.5 and O₃ in the Jiangsu and Shanghai regions (longitude range: 117.5° E–122.0° E; latitude range: 30.0° N–35.10° N) from 1 to 18 February 2019 (covering the Spring Festival period and the subsequent week). A two-layer nested model with spatial resolutions of 9 × 9 km and 3 × 3 km, respectively, was adopted for the simulation. The specific simulated region is illustrated in Fig. S4a.

The inventory developed in this study was used as the input data for traffic sources, while data from the ABaCAS database was adopted for other emission sources; these two sets of data were jointly utilized to generate the anthropogenic emission files required for the WRF-Chem model (Li et al., 2023a). The 1° × 1° Final Operational Global Analysis (FNL) data provided by the National Centers for Environmental Prediction (NCEP, https://rda.ucar.edu/datasets/ds083.2/, last access: 27 April 2026) was used to obtain the initial meteorological conditions and boundary conditions. Real-time biomass burning emissions were derived from the Fire Inventory from NCAR (FINN, https://www2.acom.ucar.edu/modeling/finn-fire-inventory-ncar, last access: 27 April 2026). The global simulation results from CAM-Chem (https://www.acom.ucar.edu/cam-chem/cam-chem.shtml, last access: 27 April 2026) were employed as the initial chemical conditions and boundary conditions. The specific parameterization schemes are presented in Table S6.

This study used observational data from 114 national monitoring stations within the simulated area to comprehensively evaluate the simulation results of the emission inventory developed in this study and the traditional inventory, as well as the improvement effects at the 15 stations located in the regions with the top 20 % road density. The simulation results were verified by calculating the normalized mean deviation (NMB) and coefficient of determination (R2).

Vehicle speed and traffic volume are important parameters for describing vehicle driving conditions and traffic flow characteristics. The nationwide average passenger vehicle speed was 42.42 km h⁻¹. The high concentration of work and business activities in the eastern China during the daytime, combined with its large vehicle population, led to generally slower vehicle speeds at 08:00 UTC+8 compared to 22:00 UTC+8 (time zone used throughout). Especially in the key cities such as Beijing, Shanghai, and Guangzhou, city vehicle speeds at 08:00 a.m. typically were 36.4 ± 0.3 km h⁻¹ due to significantly increased traffic flow during the morning peak and the circular radial road network which might concentrate traffic flow (Liu et al., 2018). During the off-peak period at 22:00, differences in traffic flow speeds across various road types were more apparent. The vehicle speeds on some provincial and county roads had increased significantly. This reflected the non-uniformity of traffic flow distribution (Guan et al., 2024).

There were also differences in average vehicle speeds between provinces, with Inner Mongolia, Jiangxi, and Qinghai having higher speeds (Fig. 1c). While Yunnan Province had the lowest average vehicle speed, which might be related to its unique topography. Approximately 94 % of Yunnan Province was mountainous, with an average elevation exceeding 2000 m (Gu et al., 2016). These complex topographical conditions influenced the actual driving speed of vehicles (Hou et al., 2019). In addition, compared to the Northeast China (NE: 08:00: 44.18 km h⁻¹; 22:00: 48.37 km h⁻¹), the average vehicle speeds in the Northwest China (NW: 08:00: 36.89 km h⁻¹; 22:00: 40.41 km h⁻¹), the East China (EC: 08:00: 39.89 km h⁻¹; 22:00: 44.56 km h⁻¹), and the South China (SC: 08:00: 40.45 km h⁻¹; 22:00: 44.16 km h⁻¹) regions were lower, due to the NE was predominantly characterized by plains, and its population and transportation network were not as dense as those in the CC and SC (Fig. 1c) (Xu et al., 2023).

Specific to the road type, the average driving speeds were higher on expressways (workday: 08:00: 67.12 km h⁻¹; 22:00: 68.59 km h⁻¹), urban arterial roads (workday: 08:00: 51.43 km h⁻¹; 22:00: 45.33 km h⁻¹) and national highways (workday: 08:00: 40.05 km h⁻¹; 22:00: 42.92 km h⁻¹) compared to other types of roads (Fig. 2a, b, c). Notably, urban arterial roads speeds fluctuated sharply during workdays and weekends, with fluctuation magnitudes of 25.3 and 12.8 km h⁻¹ respectively; this was attributable to the heavy intercity and interregional commuting traffic they carry, which caused a rapid surge in road saturation and thus a sharp speed drop, followed by a swift rebound (by 63.2 % and 41.2 % respectively) during midday off-peak hours as commuting demand wanes (Wang et al., 2016). By contrast, holiday travel was dominated by interregional family visits and tourism, with highly dispersed travel timings, leading to mitigated speed fluctuations across all road types. The emission factor of CO, VOCs, and NO_x all exhibited a trend of sharp decrease followed by gradual increase in response to vehicle speed, which was consistent with the combustion kinetics of internal combustion engines (Fig. 2d). Among them, the EFs of incomplete combustion pollutants (CO and VOCs) showed more significant responses to speed variations, while the EFs of thermal-generated NO_x were mainly controlled by the temperature and oxygen concentration inside the cylinder and thus showed relatively moderate responses to speed changes (Chen et al., 2022).

To further quantify the impact of average vehicle speed on the model results, this study conducted a quantitative assessment of model uncertainty using Monte Carlo simulation. Across all speed intervals, the emission factors and their corresponding uncertainties for CO, VOC, and NO_x were 1.4866 ± 21.42 % g km⁻¹, 0.4042 ± 22.56 % g km⁻¹, and 0.1507 ± 28.30 % g km⁻¹, respectively. Furthermore, uncertainty analysis was performed for each speed interval. Although the 40–80 km h⁻¹ interval exhibited the lowest emission factors, it represented the dominant driving range for passenger vehicles, with the highest probability density. In contrast, the uncertainty of emission factors reached its maximum when vehicle speeds exceeded 80 km h⁻¹ (Fig. S3).

Based on ride-hailing big data, the average speed of passenger vehicles fluctuated across different times. The daily average traffic speeds on workday and weekend were consistent, with both lower than on holiday, at 42.108, 42.111, and 43.032 km h⁻¹, respectively. The higher holiday speed (Fig. 3c) can be attributed to reduced urban congestion resulting from increased public transportation use for leisure travel on holidays (Zhang and Gao, 2023b). All three-day types exhibited two low-speed valleys from 07:00–09:00 and 17:00–19:00, with the workday morning valley occurring one hour earlier compared to weekend, consistent with findings by Yang et al. (2017). The national average speed frequency distribution differed at 08:00 and 22:00 on workday (Fig. 3a and b), primarily concentrated at 34.9 ± 0.2 km h⁻¹ and 44.8 ± 0.3 km h⁻¹, respectively. Compared with weekday, there were no morning and evening rush hours on holiday, resulting in a higher proportion in high-speed intervals (Yang et al., 2016). The large difference in speed between these two periods is due to the variation in traffic volumes, as illustrated by the inverse relationship between hourly traffic volume and speed distribution (Fig. 3d).

The annual amount of CO, VOCs, NO_x, PM and NH₃ emitted from national passenger vehicles in 2019 were 4087.8, 1069.4, 211.7, 1.9, 77.5 kt, respectively and TPEs was 1.6×103 kt. Considering these pollutants exhibited similar emission patterns, this study focused on analyzing the total pollutant equivalent emissions. Pollutant emissions from passenger vehicles showed significant spatial heterogeneity. Urban areas, despite occupying only 0.8 % of the country's total land area, accounted for a high 35.3 % of total vehicle emissions. High-pollution areas mainly concentrated in four urban agglomerations (BTH, YRD, PRD, SCB) (Fig. 4a). These areas were densely populated with high vehicle usage frequency, contributing significantly to the total emissions at 48.54 %. Specific emissions by province are shown in Table S7. The urban emission density of vehicles within four urban agglomerations was significantly higher compared to rural areas (Fig. 4b), as most passenger vehicles primarily operated in urban (Loder et al., 2019). However, variations in urban-rural vehicle ownership and land area across different urban agglomerations led to different urban-rural emission density gaps (Zhao and Bai, 2019). Notably, the Beijing-Tianjin-Hebei (BTH) urban agglomeration exhibited the largest urban-rural emission density ratio of 16.4, which was 13.5 higher than the national average. This gap stemmed from its higher per capita vehicle ownership than the national average, and smaller urban built-up areas than other agglomerations (Duan et al., 2024).

This study conducted a comparative analysis of the results from different studies to verify the accuracy of the emission inventory. Compared with previous research results, this research considered the impact of vehicle driving speed on emissions and the differences in emissions between holiday, workday, and weekend, and its estimated emissions in this study fell within the ranges of other estimation results (Table 1).

To improve estimation accuracy, this study obtained gridded speed-corrected emission results based on real-time vehicle driving data and compared them with passenger vehicle emission data obtained using traditional methods.

The improved methodology reflected a more accurate and detailed characterization of passenger vehicle emissions, which were primarily concentrated in urban centers, with emission intensity gradually decreasing from the city center outwards (Jing et al., 2016). Spatially, the improved method avoided the limitation of using fixed speeds in traditional approaches. It can accurately identify higher emissions on crowded urban roads caused by frequent acceleration and deceleration, and properly show lower emissions on outside roads (Fig. 7a) (Zhang et al., 2023; Wen et al., 2020; Choudhary and Gokhale, 2016). Overall, the estimation using the traditional method underestimated the total emissions by 31.5 %, with Sichuan, Beijing, Shanghai, and Guangdong being underestimated by 43.1 %, 38.4 %, 29.6 %, and 17.4 %, respectively. In Tibet, the average vehicle speed stabilized at around 42.5 km h⁻¹, and the SCC analysis revealed a speed correction factor 2.09 times that of the traditional speed correction factor (SCF), leading to the most serious underestimation in this region (Fig. 7b) (Sun et al., 2020).

The traditional method exhibited a significant underestimation of passenger vehicle emissions across distinct seasons and day types (Fig. 7c). From a seasonal perspective, this method underestimated the average daily passenger vehicle emissions by 31.6 %, 31.0 %, 32.7 % and 31.8 % in spring, summer, autumn and winter, respectively, with a relatively small overall fluctuation range. In contrast, the discrepancy in underestimation across different day types was more pronounced, and the method's underestimation of passenger vehicle emissions on weekends (33.4 %) was significantly higher than that on weekdays (27.7 %). The formation of this characteristic difference was not only associated with refined vehicle speed correction, but also stemmed from the quantitative analysis of vehicle activity levels across different day types based on congestion indices in this study.

The simulation results of the inventories reflected the temporal variation of pollutant concentrations well. The improved inventory were more consistent with the observed data from the perspective of hourly change (PM_2.5: R2=0.850, NMB = -27.4 %; O₃: R2=0.771, NMB = -32.5 %) (Figs. 8 and S8). The bias may be caused by the underestimation of the input anthropogenic emission inventory, and it falls within an acceptable and reasonable range verified by comparison with other literature (Ma et al., 2018; Wang et al., 2020; Georgiou et al., 2022). This result indicated that the accuracy of the emission inventory refined through detailed vehicle speed correction was superior to that constructed by traditional algorithms.

The inventory optimization effect was relatively notable in heavily polluted traffic-intensive areas (Fig. S4b), with an overall 0.36 % improvement in NMB (PM_2.5) and 0.02 improvement in R2 (O₃). This effect is attributable to the complex characteristics of speed fluctuations in these areas. Specifically, the NMB of PM_2.5 at the Administrative Center Monitoring Station and Putuo Monitoring Station increased by 0.7 % and 0.45 %, respectively. Meanwhile, the R2 of O₃ at the Pudong Chuansha Monitoring Station and Xinghu Garden Monitoring Station rose by 0.05 and 0.04, respectively (Fig. S9). However, this study merely achieved a slight improvement, which may be attributed to the small proportion of emissions from passenger vehicles (e.g., approximately 3.2 % for CO, 4.7 % for VOC, and 1.2 % for NO_x) (Li et al., 2023a). The performance of its simulations would be significantly enhanced when the accuracy of vehicle speed correction for the entire traffic source could be further improved.