Analysis and Prediction of PM2.5 Concentrations Over Indian Megacities Using Geospatial Techniques and Deep Learning Models

A comprehensive analysis of PM 2.5 concentrations is conducted in five major Indian metropolitan cities, namely, Delhi, Mumbai, Hyderabad, Chennai, and Kolkata. For this purpose, the required datasets are obtained from the Central Pollution Control Board (CPCB), India, and ‘Socioeconomic Data and Applications Center (SEDAC)’ of Columbia University. Spatiotemporal analysis of the SEDAC data is performed using geospatial techniques. The station observations from CPCB are used for temporal analysis and forecasting of PM 2.5 concentrations in the selected cities using four deep learning (DL) models, namely, long short-term memory (LSTM), bidirectional LSTM, gated recurrent unit (GRU), and convolutional 1D LSTM. The performance of DL models is evaluated based on the RMSE (root mean square error), MAE (mean absolute error), and R 2 (coefficient of determination), revealing that the GRU model consistently outperformed the others, displaying lower errors across all cities. Consequently, the GRU model is adopted to forecast PM 2.5 concentrations for the year 2024. A non-parametric MK (Mann-Kendall) test is conducted to analyze the trend of PM 2.5 concentrations as well. The results of the MK test provided valuable insights regarding the direction and magnitude of the trend, helping to understand the long-term behavior of PM 2.5 concentrations in these cities. The current study highlights the efficacy of geospatial techniques and DL models in analyzing and predicting PM 2.5 concentrations and underscores their importance for effective management of urban air quality.