Forecasting PM10 Concentrations in Delhi Using Time Series Analysis

The atmosphere’s particulate matter (PM) significantly contributes to climatic and meteorological shifts that severely affect all life forms. PM10 is the primary indication of the air quality index (AQI) and denotes particles with a diameter < 10 microns, the contaminant responsible for exceeding AQI criteria. Here, we use time series techniques to forecast PM10 in the Jahangirpuri region of Delhi, India, throughout the near future (a few hours to a few days in advance). Our deep learning models include Convolutional Neural Networks (CNNs), Long Short-term memory (LSTM) networks, Bidirectional LSTM (BiLSTM), Gated Recurrent unit (GRU), as well as a hybrid CNN-LSTM, CNN-BiLSTM, and CNN-GRU. We employ a multivariate time series method to anticipate PM10 concentrations and generate an outlook with uncertainty estimates. The current study compares models for forecasting PM10 concentration from measured PM2.5, SO2, NO, NO2, NOx, NH3, CO, O3, and wind speed. Training data for the models were collected from February 2018 through January 2023. Statistical measures like Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2) value are used to approximate the performance. In light of these findings, it is clear that this group of climatic parameters is crucial for estimating future levels of PM10. The study’s conclusions have important implications for policymakers, as they can use the information to inform air quality control policies and develop strategies to mitigate the adverse effects of air pollution.