This study investigated the spatial and temporal variations of PM2.5 concentrations in Harbin, China, under the influence of meteorological parameters and gaseous pollutants. The complex relationship between meteorological parameters and pollutants was explored using Pearson correlation analysis and interaction effect analysis. Using the correlation analysis and interaction analysis methods, four mechanical learning models, PCC-Is-CNN, PCC-Is-LSTM, PCC-Is-CNN-LSTM and PCC-Is-BP neural network, were developed for predicting PM2.5 concentration in different time scales by combining the long-term and short-term data with the basic mechanical learning models. The results show that the PCC-Is-CNN-LSTM model has superior prediction performance, especially when integrating short-term and long-term historical data. Meanwhile, applying the model to cities in other climatic zones, the results show that the model performs well in the Dwa climatic zone, while the prediction performance is lower in the CWa climatic zone. This suggests that although the model is well adapted in regions with a similar climate to Harbin, model performance may be limited in areas with complex climatic conditions and diverse pollutant sources. This study emphasizes the importance of considering meteorological and pollutant interactions to improve the accuracy of PM2.5 predictions, providing valuable insights into air quality management in cold regions.
Read full abstract