Unveiling the contextual nonstationary effects of the built environment on individual air pollution exposure: a mobile sensor big data driven research design.

While existing studies have extensively examined the relationship between the built environment and air pollution using monitoring station data, few have considered individual-level exposure. This study addresses that gap by adopting a real-time exposure tracking research design that integrates mobile air quality sensing and GPS tracking. Using 1.40 million mobile sensor observations, we developed an interpretable multiple XGBoost machine learning framework with SHapley Additive exPlanations (SHAP) analysis to investigate the nonstationary effects of built environment factors on individual-level air pollution exposure across indoor, outdoor, and in-vehicle contexts. The key findings are as follows. In indoor contexts, transport land use and bus stop density emerge as the most influential factors. In outdoor contexts, bus stop density, building coverage, and population density have the strongest effects. In in-vehicle contexts, population density and building coverage are the primary contributors. Besides, commercial and transport land use exhibit diminishing marginal effects. Open space, intersection density, and bus stop density exhibit threshold effects on individual exposure, with threshold values observed at 15%, 20 to 40 intersections per square kilometer, and 50 bus stops per square kilometer, respectively. Methodologically, this study introduces a transferable framework for analyzing individual-level environmental exposure in relation to the built environment. Practically, it provides a foundation for context-specific planning strategies and offers actionable guidance for prioritizing key planning indicators and selecting appropriate threshold ranges for local environmental planning.