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

Accurate assessment of anthropogenic emissions is fundamental to understanding the human-atmosphere interaction. Transportation, particularly heavy-duty trucks (HDTs), significantly contributes to air pollution (e.g., PM2.5, NOx) and global warming. The temporal and spatial variability of these emissions makes accurate inventory challenging. While emission factors for light-duty vehicles (LDVs) can be estimated with reasonable accuracy using data like vehicle registrations and fuel sales, HDT emissions are more difficult to characterize due to long-haul trips across multiple jurisdictions and frequent changes in emission standards. The rapid updates in HDT emission standards in China, Russia, South Korea, India, Turkey and Argentina, often shorter than the vehicle's lifespan, compound this challenge. Further complicating matters is the varying enforcement of emissions regulations between regions, with some areas implementing stricter standards than others. China's heterogeneous HDT emission landscape, due to rapid updates and varying policy implementation, emphasizes the large uncertainty associated with current HDT emission inventories. Improving HDT activity data is key to addressing this uncertainty, requiring a shift from aggregate proxies to individual vehicle-level descriptions. Existing methods often rely on small samples of vehicle activity data (less than 1% of the population), leading to significant errors when extrapolating to the entire fleet. Moreover, these aggregated approaches often fail to capture the impacts of sudden changes due to new policies. The study proposes a novel approach, TrackATruck, using high-volume, albeit low-precision, vehicle trajectory data to achieve full-sample enumeration for improved emission estimations.

Previous studies on vehicle emission inventories have often relied on top-down approaches using aggregate proxies such as vehicle registrations, fuel sales, and traffic volume data. While these approaches provide estimates of total emissions, they often struggle to accurately characterize the spatial and temporal distribution of emissions, particularly for heavy-duty trucks that frequently cross jurisdictional boundaries. The limitations of these aggregated methods are particularly pronounced in regions with rapidly evolving emission standards and varied implementation policies, leading to significant uncertainties in the total and distributed emissions. This study addresses these limitations by leveraging big data from vehicle tracking systems to achieve more accurate and high-resolution emission inventories.

This study utilizes the TrackATruck approach, a bottom-up methodology that uses full-sample enumeration of vehicle trajectories to construct a high-resolution emission inventory. The data source is the BeiDou Navigation Satellite System (BDS), providing over 200 billion HDT signals in the Beijing-Tianjin-Hebei (BTH) region for 2017 and 2018. TrackATruck first converts continuous BDS signals into trajectories, typically spanning 300-400 seconds. It then estimates the distribution of vehicle operating modes (Opmodes) for each trajectory based on similarity to a 1-Hz GPS trajectory database. The similarity is assessed using the speed distribution of each trajectory, selecting 1-Hz trajectories falling within the 95% confidence interval of the 1/30-Hz trajectory's speed distribution. A weighted average of the Opmode frequencies from these similar 1-Hz trajectories provides the estimated Opmode distribution for the 1/30-Hz trajectory. Emission rates for each Opmode are derived from the Ministry of Ecology and Environment of the People’s Republic of China's emission inventory (EI) guidebook and supplemented with data from a portable emission measurement system (PEMS). Finally, emission rates are combined with Opmode distributions to calculate emissions for each trajectory, which are then spatially and temporally allocated to create high-resolution emission maps and time series. The model's validation involves comparing estimated emissions from simulated Opmodes against those from observed Opmodes in a subset of five HDTs, showing a strong correlation (R²>0.8) for both NOx and PM2.5. Several existing top-down spatial allocation methods were also compared against the TrackATruck results using Pearson correlation coefficients. This comparison revealed substantial differences in spatial emission distribution between the different methods, with the TrackATruck providing a more accurate representation of the actual HDT activity patterns.

The TrackATruck model generated a high-resolution emission inventory for the BTH region, revealing significant discrepancies compared to traditional methods. Annual PM2.5 emissions from HDTs in BTH were estimated at 3739 Mg in 2017 and 3869 Mg in 2018, with NOx emissions at 136,540 Mg and 155,107 Mg, respectively. A notable 31% of NOx emissions originated from non-local HDTs. Daily variations in HDT emissions were substantial, reaching a 26-fold difference between the minimum (20.56 Mg/day) and maximum (552.31 Mg/day) NOx emissions in 2018. Spatial analysis showed high emission concentrations on intercity roads, exceeding 5 Mg/grid/year for NOx. Comparisons with existing top-down inventories revealed a 99% coefficient of variation for regional total emissions. This highlights the significant errors in existing methods, as some individual county-level emissions differed by as much as 15 times between TrackATruck and traditional methods. Even when total emission amounts were constrained to be the same, the spatial distribution differed significantly, with the top-down methods underestimating emissions along primary cargo routes by a factor of 2–10 and overestimating emissions on other routes. Temporal analysis revealed the significant impact of the Lunar New Year holiday, which dramatically reduced daily emissions. The study also analyzed the effectiveness of Beijing’s low emission zone (LEZ) policy, which showed that the overall NOx emissions remained largely unchanged while PM2.5 decreased in the LEZ but increased in upwind areas due to truck detours. The LEZ caused a 22% reduction in PM2.5 emissions within the 6th Ring Road in Beijing but only a 0.5% reduction in NOx. The 96% of emission reduction benefit from restricted HDTs was offset by increased emissions of unrestricted vehicles. The analysis of temporal allocation proxies revealed a weekday-to-weekend emission ratio of approximately 1:0.8, and the busiest hour was found to be midnight in Beijing, largely due to truck traffic restrictions during daytime. These findings strongly suggest that simple allocation schemes are inadequate for accurate spatial representation of HDT emissions, underlining the need for more sophisticated methods.

The findings demonstrate the substantial impact of incorporating high-resolution spatiotemporal data for accurate emission inventory. The TrackATruck approach significantly improves the accuracy of emission estimations compared to traditional top-down methods, particularly in representing the spatial distribution and the influence of dynamic factors such as traffic policies and holidays. The underestimation along primary cargo routes, even when total emissions are constrained, highlights the limitations of aggregate proxies in capturing the real-world driving patterns of HDTs. The LEZ policy analysis showcases the unintended consequences of localized emission control strategies, emphasizing the need for regional-level cooperative approaches that consider the movement of pollutants across borders and the broader impact on regional air quality. This study’s results suggest that a simple localized policy will result in a seesaw effect, which leads to a regional change in emissions instead of a net benefit. This calls for more integrated and comprehensive control strategies that address the demand-side of freight transportation.

This study presents TrackATruck, a novel approach for generating high-resolution HDT emission inventories using big data from vehicle trajectories. The method demonstrates significant improvements in accuracy compared to traditional approaches, revealing substantial variations in both total emissions and spatial distribution. Analysis of Beijing’s LEZ policy highlights the need for regional-level collaborative strategies to effectively mitigate the impact of HDT emissions. Future research should explore the expansion of this methodology to other regions and the development of more sophisticated spatial allocation schemes to address the limitations of simple proxies.

The study acknowledges limitations due to data availability and accuracy. The BDS data do not capture 100% of all HDTs at all times (around 30% data loss), leading to potential underestimation of emissions. Shielded signals in certain areas could further impact accuracy. Furthermore, the emission factors used were not recalibrated, potentially contributing to uncertainty. While the validation using a smaller dataset showed good correlation between simulated and actual emissions, further refinement is necessary to address these factors and improve the overall accuracy.