A big data approach to improving the vehicle emission inventory in China

Accurate estimation of anthropogenic traffic emissions is critical for understanding interactions between human activity and the atmosphere and for protecting public health. Traffic sources contribute substantially to global CO2 and NOx and to PM2.5 in Chinese megacities. Emissions vary strongly in time and space, complicating assessment and control. While improvements in activity data and emission factors can constrain total emissions for light-duty vehicles, heavy-duty trucks (HDTs) pose larger challenges due to long-haul, cross-boundary travel and heterogeneous fleets driven by rapidly updated emission standards and local advanced policies. In China, multiple HDT standards (China III–V) coexist, and regions such as Beijing implemented stricter standards earlier than neighboring provinces, producing highly heterogeneous fleets. Traditional inventories relying on aggregated proxies (e.g., registered vehicles, fuel sales, coarse VKT) struggle to represent cross-boundary HDT activity, policy shocks (e.g., low emission zones), and high-resolution distributions. This study aims to improve both total magnitude and temporal-spatial distributions of HDT emissions by leveraging full-sample vehicle trajectory big data through a new approach, TrackATruck.

Prior top-down inventories often use registered holdings, fuel sales, or limited traffic counts to estimate vehicle emissions, achieving roughly 20% uncertainty for LDVs but performing poorly for HDTs due to cross-jurisdiction travel and heterogeneous standards. Shipping inventories have successfully applied full-sample enumeration (e.g., STEAM), motivating similar methods for on-road freight. In China and other countries with rapidly tightening standards, HDT fleets contain multiple emission levels, and certain regions enforce stricter policies, leading to substantial spatial heterogeneity. Previous studies in the BTH region reported wide ranges for HDT PM2.5 (from about 2300 to 15,200 Mg per year in 2014–2015), reflecting differences in VKT, fleet composition, and emission factors. Traffic-volume-based urban studies also vary by orders of magnitude due to difficulty identifying vehicle technology categories and accounting for non-local trucks. Proxy-based spatial allocation methods (population, road density, or road density with VKT weights) have been used but may poorly correlate with actual freight activity and fail to represent hot spots like ports or policy-driven detours.

Study area and data: The Beijing-Tianjin-Hebei (BTH) region in North China (approximately 218,000 km²) was analyzed for 2017–2018. The core dataset comprises over 200 billion BeiDou Navigation Satellite System (BDS) signals for heavy-duty trucks (gross weight >12 tons) from the SINOIOV nationwide freight supervision platform, covering about 96% of medium/HDTs; QA/QC suggests roughly 70% of HDTs registered are active and present in BDS, with >80% trajectories having sampling frequencies higher than 1/30 Hz. Population data are from LandScan; road network from the National Platform for Common Geospatial Information Services; socioeconomic statistics (GDP, freight volume, new HDT registrations) from NBSC. TrackATruck framework: A bottom-up, truck-by-truck emission inventory method was developed to transform low-frequency but full-size BDS data into high-resolution emissions. The pipeline includes: (1) assembling low-frequency BDS points into trajectories of about 300–400 s (about 10 points at 1/30 Hz), ensuring continuity from the prior record; (2) estimating operating mode (Opmode) distributions for each low-frequency trajectory using the Simulated Operating Modes and Emissions (SOME) model that matches to a library of 1 Hz HDT GPS trajectories. Matching uses the speed distribution’s 95% confidence interval for the 1/30 Hz trajectory to select similar 1 Hz trajectories, then weights them by a similarity criterion to produce a simulated Opmode frequency vector across the 23 MOVES operating modes; (3) computing trajectory emissions by combining Opmode distributions with Opmode-specific emission rates and trajectory duration. Opmode emission rates are derived by mapping China’s EI guidebook emission factors and on-road PEMS data to MOVES Opmodes via a benchmark HDT model and scaling across HDT categories and standards (Pre-China I through China V, for diesel, gasoline, and others); (4) aggregating trajectory emissions temporally and spatially, allocating emissions to 0.01° grids by the time spent in each grid, and summarizing daily to annual time series and maps. Emission factors used follow the EI guidebook; NOx for China V was not reduced relative to China IV due to uncertain real-world SCR usage. Validation: Using real-world measurements from five HDTs, the SOME-based simulated Opmode distributions were compared against observed Opmodes to quantify emission estimation accuracy per trajectory. Pearson correlations between simulated and observed emissions were 0.93 for NOx and 0.87 for PM2.5; slopes were 1.04 (NOx) and 0.95 (PM2.5); mean absolute percentage errors were 9.47% (NOx) and 16.75% (PM2.5), indicating good agreement. Additional QA/QC excluded <0.01% of trajectories with missing registration or year data. Proxy comparisons: To benchmark spatial allocation, the same total HDT emissions were allocated to 0.01° grids using seven proxy schemes from prior literature (population, road density, population plus road density, and several road-density-based schemes with different VKT weights, including machine-learning optimized variants). Pearson correlations between proxy-based grids and TrackATruck grids were computed to assess spatial consistency. Policy evaluation: The method was used to assess Beijing’s low emission zone (LEZ, within the 6th Ring) implemented on 21 Sept 2017. Emissions were analyzed by standard (China III–V) and by truck registration (local vs non-local), comparing 1 Jan–20 Sept of 2017 and 2018, as well as spatial differences inside the 6th Ring and along original vs alternative intercity freight routes.

Inventory magnitudes and dynamics: In BTH, annual HDT PM2.5 emissions were 3739 Mg (2017) and 3869 Mg (2018), a 3.5% increase; NOx was 136,540 Mg (2017) and 155,107 Mg (2018), up 13.6%. Approximately 31% of NOx emissions were from non-local HDTs in both years. Day-to-day NOx varied by 26-fold in 2018, from 20.56 Mg/day (16 Feb, Lunar New Year) to 552.31 Mg/day (29 Sept). China IV HDTs contributed about 53% of PM2.5 and 55% of NOx on average across 2017–2018. Emissions were highly concentrated along major intercity roads and ring roads. Methodological comparisons: Compared with top-down studies for BTH (2014–2015), the reported PM2.5 range (≈2303–15,200 Mg/yr) differs from this study’s ≈3804 Mg/yr average by −39% to +299% (coefficient of variation about 99%), largely due to VKT, fleet composition, and treatment of non-local trucks. At the county level, using the same emission factors, 79% of counties had top-down estimates outside −50% to +150% of TrackATruck results, and 4% differed by more than 15 times. Case examples: Binhaixinqu was 4512.71 Mg/yr (top-down) vs 6835.57 Mg/yr (TrackATruck); Jingxiuqu was 5305.63 Mg/yr (top-down) vs 276.01 Mg/yr (TrackATruck). Spatial proxy schemes (population, road density with or without VKT weights) had low spatial correlations with TrackATruck grids (all below 0.5; e.g., 0.47 and 0.45 for two VKT-weighted methods), systematically underestimating emissions near ports/terminals and primary freight routes by factors of roughly 2–10 while overestimating in restricted urban cores. Temporal profiles: Updated allocation proxies show lower monthly emissions in January–February due to the Lunar New Year. Weekday-to-weekend ratio is about 1:0.8; Thursday and Friday are busiest. In Beijing, midnight is the busiest hour; night-time activity increased in 2018 relative to 2017 due to daytime truck controls. LEZ impacts (Beijing): After LEZ implementation (21 Sept 2017), emissions from banned groups (notably China III and non-local) declined, but local China V HDTs increased to fill demand. Comparing 1 Jan–20 Sept across years: NOx decreased modestly by 4% (from 6996.11 Mg in 2017 to 6729.83 Mg in 2018), while PM2.5 decreased from 179.09 Mg to 142.09 Mg. China V NOx rose from 959.48 Mg to 2980.02 Mg (3.11×) and PM2.5 from 6.81 Mg to 18.40 Mg (2.7×), about 85% due to local China V HDTs; Beijing’s new HDT registrations in 2018 were 2.7× the past 5-year average. Total HDT VKT in Beijing changed by only about +7% (3338.79 to 3594.43 million km). Inside the 6th Ring, NOx decreased only 0.5% (4416.98 to 4396.88 Mg/yr), while PM2.5 decreased 22% (110.72 to 86.39 Mg/yr). Up to 96% of non-local HDT NOx reductions were offset by increases in local HDTs. Intercity detours around the LEZ increased emissions along alternative routes by 857.72 Mg and decreased along the original route by 530.07 Mg; 79% of reductions on the original route occurred in Beijing, while increases were in neighboring upwind cities, potentially impacting regional air quality during southwest wind conditions.

The findings demonstrate that high-resolution, full-sample trajectory analysis substantially improves both the magnitude and spatiotemporal characterization of HDT emissions compared with traditional top-down proxies. Inclusion of non-local trucks is essential for local inventories, as they contribute roughly one-third of regional NOx. Proxy-based spatial allocations correlate poorly with real-world freight patterns, leading to pronounced underestimation of emissions along primary routes and near terminals and overestimation in restricted urban centers. Temporal allocation profiles derived from observed emissions offer more accurate monthly, weekly, and hourly scaling for top-down inventories. Policy analysis indicates that Beijing’s LEZ shifted emissions among vehicle classes and locations: while PM2.5 decreased in the city, NOx reductions were small due to compensatory increases in compliant local HDTs and potential limitations of real-world SCR performance. Regional detours elevated emissions in upwind neighboring cities, complicating net air quality benefits for Beijing under common meteorological conditions. These results underscore the need for regional joint control strategies and comprehensive measures, including demand-side optimization, rather than localized restrictions alone. The TrackATruck framework, tolerant of varying BDS/GPS sampling frequencies via similarity-based Opmode estimation, provides a scalable bridge from low-frequency, full-fleet data to fine-grained dynamic emissions usable for exposure assessment, hot-spot protection, and policy evaluation.

This work introduces TrackATruck, a big-data, full-sample enumeration approach that converts low-frequency, full-fleet BDS trajectories into high-resolution HDT emissions inventories. Applied to the BTH region, it reveals large day-to-day variability, significant contributions from non-local trucks, and major discrepancies between big-data and proxy-based spatial allocations. The study provides updated temporal allocation proxies and shows that simplistic spatial proxies are inadequate. Evaluation of Beijing’s LEZ highlights limited NOx benefits inside the restricted area due to fleet shifts and increased detours, reinforcing the importance of coordinated regional strategies and demand-side measures. The method is transferable to regions with widespread vehicle telematics and can support finer protection of sensitive receptors and hot spots. Future work should further refine emission factors with broader real-world PEMS datasets (e.g., SCR usage), improve coverage in challenging terrains and signal-shadowed areas, enhance detection of vehicle technology categories from telematics, and develop privacy-preserving data-sharing mechanisms to expand applicability.

The BDS trajectory dataset does not cover all HDTs at all times; QA/QC indicates about 70% of registered HDTs are active and represented, leaving possible undercoverage. Signal loss and shielding in mountainous or specific areas can lead to local underestimation. A small fraction (<0.01%) of trucks lacked complete registration or year data and were excluded. The low-frequency nature of BDS (often 1/30 Hz) requires simulated Opmode estimation, which, while validated, introduces uncertainty. Emission factors were not the focus of improvement in this study and retain uncertainty, including uncertain real-world SCR performance for China V HDTs. These limitations may affect absolute emission levels and some localized spatial estimates.