Abstract Tropospheric ozone has the potential to become an increasingly pressing public health issue in Bogotá, Colombia, due to rising concentrations across the city driven by complex interactions among emissions, meteorology, and urban structure. This study presents a comprehensive spatiotemporal analysis of ozone levels from 2013 to 2023 and assesses the associated health burden using mortality data from the same period. Results reveal a consistent upward trend in ozone concentrations, particularly in northern, western, and southern localities, with seasonal peaks linked to biomass burning and photochemical conditions. Mortality analysis, based on the Global Exposure Mortality Model, estimates that 18. 3% of all deaths among individuals aged 25 and older are attributable to long-term ozone exposure. The highest burdens are found in densely populated and socioeconomically vulnerable areas such as Kennedy, Suba, and Ciudad Bolívar, with the elderly being the most affected. Building on these findings, we developed a machine learning prediction model for ozone using a convolutional merge with a long-short term memory network architecture trained on air quality and meteorological variables. The model demonstrated strong predictive performance (mean Rho=0. 86, RMSE=3. 5 g/m 3) across monitoring stations (17 with at least 35000 data points), supporting its potential application in real-time early warning systems across Bogotá. This integrated approach highlights the importance of localized air quality management, combining epidemiological assessment with predictive modeling. The findings underscore the urgency of implementing region-specific mitigation strategies and improving monitoring infrastructure to reduce health risks from ozone exposure in Bogotá’s rapidly growing urban environment. Graphical Abstract The top-left panel shows a rising trend in tropospheric ozone concentrations across Bogotá, highlighting the increasing severity of air pollution. The top-right panel represents the associated health burden, indicating elevated mortality risks, particularly among older and vulnerable populations. The bottom-left panel depicts the machine learning framework developed in this study, integrating a convolutional layer and a bidirectional long-short term memory architecture to predict future ozone levels with high accuracy, showing its potential as a base for early warning systems. Finally, the bottom-right panel emphasizes the study’s geographical focus on Bogotá, Colombia.
Bustos et al. (Fri,) studied this question.