Integrating Fragmented Mobile Monitoring Data into a High-Resolution Spatiotemporal Exposure Model for Black Carbon

Particulate black carbon (BC) is a critical traffic-related air pollutant associated with adverse health effects in susceptible populations, but exposure modeling remains challenging because BC is not routinely monitored in regulatory networks. This study developed a mobile monitoring-based spatiotemporal exposure modeling framework for estimating ambient BC exposure in a pregnancy cohort in Beijing, China. Equivalent black carbon (eBC) was measured using a vehicle platform during three seasonal campaigns from 2023 to 2024 in urban residential areas, generating a spatially rich but temporally fragmented real-time data set that was aggregated to the hourly level. To fill temporal gaps at locations with mobile BC measurements, we integrated mobile measurements with external temporal predictors using tree-based machine learning models and interpreted feature contributions using SHapley Additive exPlanations (SHAP) values. The resulting site-level daily BC estimates were subsequently used to develop a hierarchical geostatistical model based on geographic covariates, which was applied directly to estimate daily BC exposure at unmonitored cohort locations. The eXtreme Gradient Boosting (XGBoost) model achieved a 10-fold cross-validation mean squared error-based R2 (Rmse2) of 0.69 against measured site-hour eBC, while the subsequent spatiotemporal model showed strong internal interpolation performance for the XGBoost-derived daily BC estimates at the mobile monitoring locations. This study provides a practical framework for integrating temporally fragmented mobile monitoring data with external data sets to support BC exposure assessment at high spatial resolution in subsequent epidemiological studies and health risk assessment.