Integrating Long-Term Mobile Measurements with Geospatial Modeling for On-Road PM2.5 Assessment in Hong Kong

Fixed monitoring networks characterize urban PM2.5 trends but cannot resolve variation among individual roads and transport corridors. This study developed and evaluated a two-stage framework integrating long-term mobile measurements with satellite-assisted geospatial modeling to estimate on-road PM2.5 at 30 m resolution in Hong Kong during 2018–2025. The first-stage TOA-LUR-GTWR model used regulatory observations, Landsat 8/9 and Sentinel-2 top-of-atmosphere (TOA) reflectance, meteorology, and geographic predictors. Quality-controlled AirCasting measurements were matched to roads and aggregated into 1184 road-segment-date observations. Mobile-minus-base residuals were modeled using a Light Gradient Boosting Machine (LightGBM) and evaluated on held-out roads and against independent roadside regulatory observations. The mobile-informed estimates, defined as base estimates adjusted using residuals learned from mobile observations, increased R2 from 0.73 to 0.85 and reduced RMSE from 12.31 to 9.02 µg/m3 and MAE from 8.68 to 6.42 µg/m3. Major road density, sky view factor, wind speed, Green View Index, and built-up density were the leading residual predictors. Annual concentrations generally declined after 2018, although road-level contrasts persisted. These results indicate that long-term mobile measurements can capture road-scale PM2.5 variation not resolved by satellite-assisted base estimates alone and extend fine-scale estimation beyond directly sampled routes.