High-resolution PM2.5 mapping and evaluation in Tehran using mobile monitoring, land-use regression, and explainable hybrid models
Year: 2026
Type: article
Abstract: Urban PM 2.5 modeling has traditionally relied on satellite-derived estimates and sparse fixed-site monitoring data. Although these sources provide broad spatial coverage, their accuracy is often limited by cloud interference and surface reflectance, reducing their reliability in capturing near-surface exposure in densely urbanized environments. Therefore, ground-based monitoring has become essential for resolving intra-urban air quality patterns and informing mitigation strategies. This study presents one of the most comprehensive mobile PM 2.5 exposure assessments in Tehran by combining year-long mobile monitoring with data-driven modeling. Across four seasonal campaigns (June 2023–March 2024), five pedestrian routes representing diverse land-use contexts were monitored using co-located low-cost sensors (SEN55, SPS30, AirBeam3), yielding over 380 h of high-frequency data spanning more than 550 km. These data were used to characterize spatiotemporal exposure patterns and train predictive models using Land-Use Regression, Random Forest, and XGBoost. The predictors included in situ meteorological variables and geospatial land-use features extracted from buffers ranging from 50 to 500 m. Multiple input configurations, including unscaled, scaled, meteorology-only, and statistically selected configurations, were evaluated. RF achieved the highest predictive performance (R 2 = 0.89), followed by XGBoost (R 2 = 0.88), whereas LUR performed moderately (R 2 = 0.66). Model interpretability was enhanced using SHAP, which identified the key predictors contributing to the model outputs. The best-performing models were used to generate high-resolution PM 2.5 maps for summer and winter, which are the city's most critical air quality seasons. The results demonstrate the utility of integrated mobile monitoring and hybrid modeling for scalable, interpretable, and high-fidelity urban air quality assessments. • Year-long mobile PM 2.5 monitoring using co-located low-cost sensors in Tehran. • Hybrid models (RF, XGBoost) outperform LUR in predicting urban PM 2.5 levels. • Multiple data configurations evaluated for optimal model input structure. • Fine-scale PM 2.5 maps and dose metrics reveal seasonal pedestrian risks. • SHAP analysis enabled interpretation of predictor importance in hybrid models.
Source: Urban Climate
Authors: Alireza Shakerdonyavi, Bijan Yeganeh
Institution: Shahid Beheshti University
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,690
OpenAlex ID: W7123837960