Estimation of daily surface ozone using periodic and stochastic modeling in Chennai region

Abstract

The present study deals with the modeling and forecasting of surface ozone time series in an urban area. First, an analysis of the systematic components (periodicity and stochastic components) was performed. Subsequently, prediction model for the daily surface ozone series was developed. In the recent years, there was no permanent measurement of surface ozone at this site, so surface ozone was measured from June 2011 to September 2012 at the urban site Koyembedu, Chennai (the capital of Tamil Nadu), India. Daily cumulative ozone data series was obtained by hourly instantaneous data. It was found, using Mann-Kendall test, that the data series is free of trend. The periodicity of ozone data was analyzed using Fourier Transform method. Stochastic components of ozone data are assumed as residues between observed ozone data and values computed from periodic model. Stochastic model presented in this research is basically a 3rd order autoregressive model. The developed models were validated using correlation coefficient between the predicted values and measured values. The spectrums of series exhibit 100 days period of daily surface ozone and implies that the pattern in the series is repeated every 100 days. The correlation coefficient (R) of this model delivers 0.810 and can provide mean bias error (MBE) = 0.85, and root mean square error (RMSE)=0.83. The result suggests that this approach is good for estimating daily surface ozone with sufficient accuracy.