<p>Tehran, the capital of Iran and one of its most polluted megacities, has over 9 million residents who regularly experience periods of severe pollution. This study addresses four key challenges in air pollution modeling: the non-normality of data, spatial correlations between data points, misalignment between meteorological and air pollution datasets, and misalignment between air pollution and population data. To tackle these issues, we applied a transformed Gaussian spatial model to manage non-normality. A Bayesian framework was then employed for spatial prediction, leveraging Monte Carlo methods to sample the posterior distribution. To resolve the misalignment issues, we developed a comprehensive grid network of air quality monitoring stations. This network enabled precise prediction of meteorological parameters within each grid segment, and the same approach was extended to accurately estimate population exposure. Our findings demonstrate that the Bayesian Transformed Gaussian (BTG) model significantly outperforms traditional Gaussian and Log-Gaussian models in terms of fitting accuracy, predictive performance, and consistency with external datasets when analyzing air pollutant concentrations in Tehran. The study identified the most polluted regions as the western area on warm day (with higher wind speed and visibility) and the central area on cold day. On warm day, 1.1 million people were exposed to the highest O₃ levels, while on cold day, 1.6 million people experienced the maximum PM_2.5 concentrations. A comparison between modeled and satellite data within the grid networks revealed satisfactory alignment (R² &gt; 0.5) on cold day. Notably, field data-based model results were more consistent with satellite measurements on warm day. This study provides valuable insights into selecting effective and accurate methods for determining air pollutant concentrations in specific regions.</p> Graphical Abstract <p></p>