Abstract
Although fine particles (PM
2.5
) concentrations in China have declined over the past decade, exposure disparities persist at hyperlocal scales. We conducted a six-month mobile monitoring campaign in Shenzhen, where four taxis with calibrated low-cost sensors gathered 24.6 million 1 Hz measurements (~3,700 hours) across 550 km² and 7.3 million residents. We developed a novel application of a multi-source XGBoost (MS-XGB) framework that integrates mobile and fixed-site station data to reconstruct a road-network PM
2.5
dataset at 30 m × 1 h resolution, filling 98% of missing observations. Validation demonstrates strong agreement with both mobile observations (R
2
= 0.97) and data from fixed regulatory monitoring stations (Pearson r = 0.82). We used the road-network dataset to generate a 100 m × 100 m population grid exposure map and benchmarked it against two conventional exposure assessment approaches that rely solely on fixed-site regulatory stations. Relative to MS-XGB, conventional methods systematically underestimated long-term PM
2.5
exposure, with grid-level differences of up to 15 μg/m³. Population-weighted mean exposure was higher in urban villages (16.51 µg/m³) than in other urban areas (14.41 µg/m³), with hotspots 52.5% more likely to occur in urban villages. In Longhua, high exposure was concentrated specifically in urban villages, indicating that district-level averages can mask substantial intra-district heterogeneity. In Nanshan, 40% of elderly residents live in high-exposure areas, compared with 34% of children aged 0–14 and 33% of adults aged 15–59, suggesting that older adults bear a disproportionate exposure burden and face elevated health risks. Our findings demonstrate that mobile-monitoring-based hyperlocal exposure assessment is essential for revealing true PM
2.5
exposure inequalities across population groups and spatial environments that are overlooked by conventional fixed-site approaches.