Implementing Spectral Decomposition of Time Series Data in Artificial Neural Networks to Predict Air Pollutant Concentrations

A model to predict air pollutants' concentrations was developed by implementing spectral decomposition of time series data, obtained by Kolmogorov-Zurbenko filter, in Artificial Neural Networks (ANN). This model was utilized to separate and individually predict three spectral components of air pollutants' time series of short, seasonal, and long-term. The best set of input variable was selected by evaluating the significance of different input variables while modeling different time series components. Moreover, different possible approaches for constructing such models were examined. Performance of the constructed model to predict air pollutants' level at a central location in Tehran, Iran, which is one of the most polluted cities in the world, was assessed. The constructed model showed firm and reliable performance in modeling and predicting the two selected air pollutants of NOx and PM10. The R-2 between predicted and observed values were similar to 0.90 for most cases. It was shown that the developed model could perform better in modeling air pollutants compared with ordinary ANN models, especially in episodes of highly elevated pollution levels. Furthermore, this model provided the opportunity to separately predict pollutants' spectral components, such as baseline concentrations, which represent urban background levels. Predictions of baseline concentrations were also in fine agreement with the observed data. Such modeling and prediction could help policymakers to oversee different trends of pollutants' fluctuations, and make proper decisions to control the pollutants.