Assessment of ground level ozone transformational behaviour during critical conversion time using parametrical and non-parametrical analysis

Ground Level Ozone (O3) is the second most significant air pollution in Malaysia and could cause negative impact towards human health. This study aimed to investigate the O3 transformational behaviour during a duration defined as Critical Conversion Time (CCT). The CCT is a period where the minimum to maximum rate of O3 transformation occurs. This study utilized O3 hourly average secondary monitoring records over 12 years (2000 to 2011) at 11 locations in Peninsular Malaysia with different land use settings i.e. urban, sub-urban, industrial and background. The parameters used in this study comprises O3, its precursors (NO2, NO, CO, CH4, NMHC and SO2), and meteorological parameters (UVB, wind speed, wind direction, temperature and relative humidity). The variation of O3 were analysed using descriptive statistics; time series analysis is then used to overview the O3 trend and pattern as well as to identity exceedances in the data. The NO-NO2-O3 relationship analysis was carried out using composite diurnal plot. The diurnal plots were also used to determine the CCT period based on the critical conversion point (CCP). CCP is the point where the diurnal line of O3 intercepts with diurnal line of NO and NO2 during morning build-up period, while the CCT is defined as the period where minimum to maximum time for CCP. Next, O3 prediction models during CCT were developed using two analyses which are Multivariate Linear Regression (MLR) as the parametrical analysis and Partial Least Square (PLS) for non-parametrical analysis. The developed models were constructed using MLR models that have potential impact on the covariates, therefore this study tried to enhance the O3 prediction using non-parametrical PLS models to avoid the autocorrelation and multicollinearity problems. The developed models performances in terms of accuracy and error were verified and validated using performance indicators namely Prediction Accuracy (PA), Coefficient of Determination (R2) and Index of Agreement (IA), Normalized Absolute Error (NAE), Root Means Square Error (RMSE) and Mean Square Error (MSE). The model validation used 20% randomly selected secondary data, while the verification were carried out using primary data that were collected at three locations which are Shah Alam (urban), Bakar Arang (Sub Urban) and Jerantut (Background). Results suggested that O3 in the selected study areas showed a typical unimodal trend with CCT period is identified to occur earliest at 8 a.m. (Kota Bharu, Ipoh, Bakar Arang and Jerantut) and latest at 11 a.m. (Shah Alam, Kajang, Johor Bahru, Nilai, Seberang Prai and Pasir Gudang). So, the period of 8.00 a.m. to 11 a.m (4 hours) was used as CCT in further analysis of this study. Results showed that MLR models exhibited significantly higher performance compared to PLS models in all stations based verification and validation performances except for verification in Jerantut which showed the opposite result. The PLS verification model in Jerantut excelled in two out of three accuracy performance with values of R2 (0.852) and PA (0.963), otherwise MLR model showed higher IA (0.721) with relatively lower error as indicated by RMSE (13.481), MAE (10.750) and NAE (0.509). The obtained results demonstrated that the CCT period model that was compared using MLR and PLS analyses can be utilised. The usage of PLS model during CCT period model in estimating the daytime ozone concentration could be an alternative for ozone predictions especially in background areas.