[Inverse Modeling of PM_2.5 Mass Concentration Using a Particle Swarm Optimization XGBoost Model]

As a major source of air pollution, PM_2.5 poses a serious threat to human health, and the study of PM_2.5 concentration estimation methods helps to identify pollution sources, optimize air quality management measures, and effectively improve the quality of the urban environment. To obtain the spatial distribution of PM_2.5 mass concentration over a large range and improve the accuracy of PM_2.5 estimation, a model of particle swarm optimized XGBoost （PSO-XGB） was proposed, and a particle swarm optimization algorithm （PSO） was introduced to optimize the parameters of XGBoost. By integrating the aerosol optical depth product （AOD） from the moderate resolution imaging spectroradiometer （MODIS） and meteorological data, the PSO-XGB-based PM_2.5 mass concentration inversion model was constructed, followed by inversion of the national PM_2.5 mass concentration data in 2022 by year and season, and the PSO-XGB model performance was evaluated using the ten-fold cross-validation method. Finally, the national PM_2.5 mass concentration data of seven years and only one year were used to input into the model for training, respectively, to analyze the effect of the amount of data on model performance. The results showed that the PSO-XGB model was able to effectively invert PM_2.5 mass concentration, and the overall R² reached more than 0.9, of which the best inversion effect was 0.98 in winter, 0.96 in autumn, 0.90 in summer, and 0.89 in spring. Simultaneously, the comparative analysis showed that the amount of data had no effect on the inversion performance of the optimization model, while dividing the data into appropriate chunks by time series, or dividing seasons, could more accurately assess the stability and applicability of the performance of the model.