Spatiotemporal Analysis and Deep Learning-Based Prediction of Air Pollution in China, 2015–2024

Air quality in China has markedly improved over the past decade, yet pollution levels remain high and continue to threaten public health. This study analyzed the spatiotemporal variations in six air pollutants (PM2.5, PM10, SO2, NO2, O3, CO) across seven regions in China (2015–2024) using Kriging interpolation. The performance of Recurrent Neural Network (RNN), Long Short-term Memory (LSTM), and Convolutional Neural Network-Long Short-term Memory (CNN-LSTM) models was assessed using the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2) metrics. Results showed that all pollutants exhibited overall declining trends, with SO2 depicting the largest reduction (69.53%), while O3 displayed intermittent increases from 2017 to 2024. North China recorded both the highest concentrations and the greatest reductions in PM2.5, SO2, and CO, whereas Southwest and South China maintained the lowest overall levels. Among the predictive models, LSTM achieved the highest overall accuracy (mean RMSE = 1.802, mean MAE = 0.915, R2 > 0.99). These findings provide a comprehensive depiction of China’s air pollution evolution and highlight the potential of deep learning for region-specific air quality prediction and policy design. The results offer a quantitative foundation for optimizing differentiated control strategies and advancing precision air quality management.