Development of machine learning models for estimating ambient air pollutants (PM2.5, and O3) concentrations and health risk assessment across cold and warm seasons in 22 districts of Tehran megacity

Air pollution, especially in Asia, poses a serious threat to public health and the environment, highlighting the need for accurate urban air pollution prediction models for better management and health protection. However, previous studies have often paid limited attention to fine-scale urban variations and the integrated assessment of air pollution and associated health risks. This study develops machine learning models to estimate PM 2.5 and O 3 concentrations in Tehran during warm and cold seasons by integrating meteorological parameters with satellite-derived indices. Following the collection of data on the target pollutants, meteorological parameters, and satellite indices, machine learning algorithms utilizing 20 predictor variables were trained and evaluated. The results showed that PM 2.5 and O 3 exhibited contrasting seasonal and daily patterns. PM 2.5 concentrations were higher during the cold seasons and peak traffic hours, whereas O 3 concentrations reached their highest levels during the warm seasons and afternoons due to sunlight-dependent photochemical reactions. For predicting PM 2.5 concentrations, the Extra Trees algorithm performed best during cold seasons (R 2 =0.6), while the CatBoost model achieved higher accuracy during warm seasons (R 2 =0.7). Meanwhile, XGBoost was consistently identified as the superior model for predicting O 3 concentrations across both cold and warm seasons, achieving an R 2 of 0.7 in both periods. The PM 2.5 risk assessment showed that HQ values were generally higher during cold seasons than warm seasons, and exceeded the reference level of 1 across all regions. The non-carcinogenic health risks associated with PM 2.5 inhalation were assessed using the U.S. EPA methodology, with Monte Carlo simulation used to characterize variability and uncertainty. These findings provide scientific evidence to support targeted air quality management and health risk mitigation in urban environments.