Heterogeneous causal effects of mobile source emissions on ozone: A multi-source big data fusion and double machine learning approach

DOI number:10.1109/TBDATA.2026.3728704

Teaching and Research Group:TURBO LAB - 交通与城市运行实验室

Journal:IEEE Transactions on Big Data

Abstract:Urban ozone pollution is strongly influenced by traffic emissions, but the conditional effects of different vehicle-related proxies and their interactions with atmospheric conditions remain insufficiently understood. This study addresses this gap by proposing a double machine learning (DML) framework to estimate nonlinear and heterogeneous traffic-related effects under explicit identifying assumptions, focusing on general traffic and heavy-duty diesel vehicles in Chengdu, China. Using GPS data, congestion indices, meteorological parameters, and multi-source adjustment covariates, a DML approach combining random forest and Lasso regression with cross-fitting is employed to obtain orthogonalized conditional effect estimates. Results indicate that while general traffic proxies are associated with ozone variability, construction transport vehicles, though fewer in number, show disproportionate estimated effects under specific atmospheric conditions. These findings highlight the need to integrate meteorological variability and source-specific contributions into urban air quality management, enabling more targeted and effective emission control policies.

Indexed by:Journal articles

Translation or Not:no

Date of Publication:2026-08-28

Included Journals:SCI

Links to published journals:https://ieeexplore.ieee.org/document/11669972