Vehicle-based monitoring and AI unravel patterns of on-road carbon and pollutant emissions

Achieving sustainable urban development necessitates a significant reduction in carbon dioxide (CO2) emissions from transportation. Urban road traffic CO2 concentrations display intricate spatial patterns influenced by street layouts, mobile sources, and human activities. However, a comprehensive grasp of these patterns, which entail complex interactions, remains elusive due to the omission of human perspectives on road interface characteristics.

Our research team has developed an innovative integrated AI carbon emission monitoring technology through vehicle-based surveys. This technology utilizes panoramic visual sensors and various greenhouse gas (GHG) analyzers for spatiotemporal collaborative observations, data processing, and modeling. It provides insights into the dynamic connections between the physical urban space and road traffic emissions, offering a precise and refined carbon and pollutant emission source tracing system. This method automatically extracts attributes of objects and landscapes in urban scenes, aiding in evaluating the relative importance of built environments and road traffic to emission intensities in real scenarios. Based on a thorough understanding of in-situ conditions, this approach aims to identify coordinated development paths for buildings and transportation to enhance emission reduction effects.

In this study, a mobile travel platform was constructed to collect on-road navigation Street View Panoramas (OSVPs) and corresponding CO2 concentrations, obtaining over 100,000 sample pairs covering 675.8 km of roads in Shenzhen, China. Four ensemble learning (EL) models were used to establish nonlinear connections between the semantic and object features of streetscapes and CO2 concentrations. After EL fusion modeling, the predictive R2 in the test set exceeded 90%, and the mean absolute error (MAE) was <3.2 ppm. The model was applied to Baidu Street View Panoramas (BSVPs) in Shenzhen to generate a 100 m resolution map of average on-road CO2, and the Local Indicator of Spatial Association (LISA) was used to identify high CO2 intensity spatial clusters. Light Gradient Boost-SHapley Additive exPlanation (LGB-SHAP) analysis revealed that vertically planted trees can reduce on-road CO2 emissions. Moreover, factors affecting on-road CO2 exhibit interaction and threshold effects. Street View Panoramas (SVPs) and Artificial Intelligence (AI) were used to enhance the spatial measurement of on-road CO2 concentrations and the understanding of driving factors. This approach facilitates the assessment and design of low-emission transportation in urban areas, which is critical for promoting sustainable traffic development.