Construction of a Model for Estimating PM2.5 Concentration in the Yangtze River Delta Urban Agglomeration Based on Missing Value Interpolation of Satellite AOD Data and a Machine Learning Algorithm

Air pollution is an important environmental issue that affects social development and human life. Atmospheric fine particulate matter (PM2.5) is the primary pollutant affecting the air quality of most cities in the authors’ country. It can cause severe haze, reduce air visibility and cleanliness, and affect people’s daily lives and health. Therefore, it has become a primary research object. Ground monitoring and satellite remote sensing are currently the main ways to obtain PM2.5 data. Satellite remote sensing technology has the advantages of macro-scale, dynamic, and real-time functioning, which can make up for the limitations of the uneven distribution and high cost of ground monitoring stations. Therefore, it provides an effective means to establish a mathematical model—based on atmospheric aerosol optical thickness data obtained through satellite remote sensing and PM2.5 concentration data measured by ground monitoring stations—in order to estimate the PM2.5 concentration and temporal and spatial distribution. This study takes the Yangtze River Delta region as the research area. Based on the measured PM2.5 concentration data obtained from 184 ground monitoring stations in 2023, the newly released sixth version of the MODIS aerosol optical depth product obtained via the US Terra and Aqua satellites is used as the main prediction factor. Dark-pixel AOD data with a 3 km resolution and dark-blue AOD data with a 10 km resolution are combined with the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis meteorological, land use, road network, and population density data and other auxiliary prediction factors, and XGBoost and LSTM models are used to achieve high-precision estimation of the spatiotemporal changes in PM2.5 concentration in the Yangtze River Delta region.