Fusing satellite imagery and ground-based observations for PM_2.5 air pollution modeling in Iran using a deep learning...

With the rapid advancement of urbanization and industrialization in cities, air pollution has become one of the significant environmental challenges and issues in many countries. The concentration of particulate matter with an aerodynamic diameter of less than 2.5 µm (PM_2.5), recognized as the main air pollutant in Iran, penetrates the respiratory system through inhalation, causing respiratory and cardiovascular diseases, reproductive disorders, central nervous system disturbances, and cancer. Accurate and high spatiotemporal modeling of air pollutant concentrations is crucial for air quality management and exposure assessment in epidemiological studies. Air quality monitoring stations provide valuable data on air pollution levels at specific locations, but they have limitations in fully capturing the air quality across entire areas and cities in a country. Meanwhile, the rapid growth of computational technologies and the availability of air quality data have enabled researchers to develop complex models using deep learning for modeling various air pollutant concentrations. In this research, PM_2.5 pollutant concentration modeling for monthly continuous distribution estimation has been implemented and evaluated based on deep learning models. The results of the presented models were compared to select the model with the best performance. Additionally, the influence of the parameters used on pollutant concentration levels was analyzed. In this study, we applied deep learning techniques, including Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Convolutional Long Short-Term Memory (ConvLSTM), to model PM_2.5 concentrations for continuous monthly distribution estimation. We utilized satellite data, ground-based observations, and meteorological parameters as input features. The models were evaluated using Root Mean Square Error (RMSE) and the coefficient of determination (R²). The ConvLSTM model outperformed others with an RMSE of 4.95 µg/m³ and an R² of 91.24%. Sensitivity analysis indicated that among 18 input parameters, population density, AOD, NO₂, and rainfall had the most significant impact on PM_2.5 concentrations.