Nonlinear Dynamical Characterization and Prediction of Ambient Nitrogen Dioxide Concentration

Air quality predictions in an area are required in order to enable the concerned authorities to issue the warning against the air pollution episodes. We propose the method to predict the air quality based on the neural networks and nonlinear dynamical systems theory, which is helpful if the data on the explanatory variables influencing the time series is not available. Using this concept, Nitrogen dioxide (NO2) concentration at three sites in Kolkata, India was characterized for invariant measures to ascertain the presence of chaos in the time series. The phase space was reconstructed to estimate the invariant measures such as correlation dimension. It is observed that the process generating NO2 time series is deterministic. The predictions were obtained using the neural network model with backpropagation training. The model evaluation results indicated that if the appropriate value of embedding dimension is given to the neural network, it is capable of predicting the chaotic time series of NO2 concentration.