Untitled

Abstract

The present study includes prediction of next day hourly ozone concentration using four models viz. multiple linear regression (MLR), principal component regression (PCR), artificial neural network (ANN) and principal component based artificial neural network (PCANN). The input variables used for models construction were hourly concentration of previous day ozone, nitrogen dioxide (NO 2 ), carbon monoxide (CO), temperature (T), relative humidity (RH), wind speed (WS), solar radiation (SR) and solar radiation duration (SRD). The measurement of ozone and its precursors was carried out at a semi-urban site of Dayalbagh, Agra. The models showed good agreement with observed levels of ozone. The value of regression coefficient ranged from 0.85 to 0.92 for different models. The highest value of regression coefficient was observed for PCANN model. In addition, normalized absolute error (NAE), root mean square error (RMSE), index of agreement (IA) and mean biased error (MBE) were also calculated to check the performance of different models. The principal component-based ANN model was the best model as it is associated with maximum value of regression coefficient (R = 0.92) and minimum value of errors. The efficiency of models was also checked for unknown datasets that were not used in model construction.