Assessing relative population exposure and non-carcinogenic health risks of PM2.5 in Tehran through machine learning, satellite observations, and uncertainty analysis

• A seasonal GBM model was used to estimate long-term PM 2.5 in Tehran. • Model performance showed strong agreement with observations (R = 0.74–0.87). • Central and eastern districts experienced higher PM 2.5 exposure and health risks. • Health risk estimates were highly sensitive to the selected reference concentration. Fine particulate matter with an aerodynamic diameter below 2.5 µm (PM 2.5 ) is recognized as a major urban air pollutant, closely associated with adverse health outcomes. Despite the well-established risks of PM 2.5 , accurately estimating its spatial distribution and population exposure remains a major challenge due to the heterogeneity of emission sources and complex atmospheric patterns. Therefore, this research aimed to develop a simple yet effective framework for precisely estimating the spatial distribution of PM 2.5 in Tehran and evaluating its associated health impacts. First, daily aerosol optical depth (AOD) data at a spatial resolution of 1 km were obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) remote sensing dataset. Afterward, seasonal gradient boosting machine (GBM) models were used to integrate AOD data and meteorological variables with ground PM 2.5 observations, to allow for spatiotemporal estimation of this pollutant between 2010 and 2021. The validation step demonstrated satisfactory performance of the GBM models for PM 2.5 estimation, with correlation coefficients ranging from 0.74 to 0.87. Subsequently, the PM 2.5 estimates were combined with population density to evaluate population exposure using the relative risk index (Q i ) across the 22 municipal districts of Tehran, followed by an assessment of non-carcinogenic health risks. The Relative Exposure Risk (RER) analysis indicated that central areas of Tehran experienced higher risk levels, while western districts were comparatively safer. Estimated PM 2.5 concentrations were evaluated for non-carcinogenic health risks using the hazard quotient (HQ) approach, where HQ = 1 represents the threshold of concern. Under the regulatory benchmark (reference concentration (RfC) = 35 µg/m 3 ), mean HQs across Tehran remained below unity (0.84–0.98), though several central districts approached this limit. Applying health-based criteria substantially increased risk estimates, with mean HQs of ∼2.0–2.4 (RfC = 15 µg/m 3 ) and ∼6.1–7.1 (RfC = 5 µg/m 3 ). Monte Carlo uncertainty analysis (10,000 iterations) further revealed that the probability of HQ exceeding unity exceeded ∼94–97% under WHO-based benchmarks and approached 100% under the most stringent scenario, while remaining non-negligible (∼29–43%) even under the regulatory threshold. Overall, PM 2.5 -related health risks were highly sensitive to the selected reference concentration, with consistently elevated risks in central and eastern Tehran.