Combining principal component regression and artificial neural networks for more accurate predictions of ground-level ozone

This work encompasses ozone modeling in the lower atmosphere. Data on seven environmental pollutant concentrations (CH4, NMHC, CO, CO2, NO, NO2, and SO2) and five meteorological variables (wind speed, wind direction, air temperature, relative humidity, and solar radiation) were used to develop models to predict the concentration of ozone in Kuwait's lower atmosphere. The models were developed by using summer air quality and meteorological data from a typical urban site when ozone concentration levels were the highest. The site was selected to represent a typical residential area with high traffic influences. The combined method, which is based on using both multiple regression combined with principal component analysis (PCR) and artificial neural network (ANN) modeling, was used to predict ozone concentration levels in the lower atmosphere. This combined approach was used to improve the prediction accuracy of ozone. The predictions of the models were found to be consistent with observed values. The R2 values were 0.965, 0.986, and 0.995 for PCR, ANN, and the combined model prediction, respectively. It was found that combining the predictions from the PCR and ANN models reduced the root mean square errors (RMSE) of ozone concentrations. It is clear that combining predictions generated by different methods could improve the accuracy and provide a prediction that is superior to a single model prediction.