| Summary: | This research involves the analyses of secondary air quality data collected at 10 monitoring stations in Malaysia started from January 2006 until December 2012. The objectives of this study are to assess the air quality status in studied area in Malaysia, identify the most significant variables in discriminating the spatial pattern of air quality based on the historical data, determine the source apportionment of air pollution, and establish the input-output relationship based on data driven modelling of API in the study area. Descriptive statistical analysis was performed to compare the status of Malaysian air quality with the Recommended Malaysian Air Quality Guidelines (RMAQG). Several environmetric techniques were applied to this seven year daily average database. Three types of environmetric techniques were chosen: hierarchical agglomerative cluster analysis (HACA) to access the spatial air quality patterns, discriminant analysis (DA) to investigate the significant discriminating air quality variables, and principal component analysis (PCA) to determine the probable sources of air pollutants in the study area. The artificial neural networks (ANN) technique was applied to develop several models. The spatial distribution model was applied to investigate the potential use of selected variable parameter obtained by the DA for the purpose of discriminating that parameters on the three spatial regions (Higher Polluted Area, HPA; Moderate Polluted Area, MPA; and Lower Polluted Area, LPA) obtained by HACA. A sensitivity analysis model was developed to identify the most significant variables in Malaysia using leave-one-out method. The receptor model (PCA/ANN) was applied to identify the source apportionment of the significant variables and estimate the source contributions in the atmosphere. Besides, the ANN Model with different input parameters have been applied to predict the air quality status in 2020. The descriptive statistics computed show that the mean values of all the parameters are within the value of RMAQG. HACA grouped the 10 air monitoring stations into three different clusters known as the HPA, MPA, and LPA. Forward and backward stepwise DA manage to discriminate seven (64.90% correct ) and five (66.12% correct) air quality variables, respectively, from the original of eight variables (64.83% correct). The spatial distribution model showed better prediction performance in discriminating the regions, with an excellent percentage of correct classification was 77.66% compared to the DA method. The sensitivity analysis model explained that from eight parameters, only five parameters (PM, CO, CH, THC, and NmHC) were categorized as the most significant for determining the status of air quality in Malaysia. In addition, the findings from sensitivity analysis showed that more than 80% of Air Pollutant Index (API) values were influenced by PM and CO. The receptor model showed better predictive ability in determining of the API uses fewer variables, with the R and RMSE values of 0.618 and 10.017 (p < 0.05), respectively. Different input parameters will give different R and RMSE values, where based on the three models developed, the value of R and RMSE of ANN Model 1, ANN Model 2, and ANN Model 3 were 0.7261 and 6.1825, 0.6904 and 6.7180 and 0.6649 and 6.9876, respectively. Hence, it can be concluded that the ANN Model 1 with parameters of NO, PM, CH, CO and THC were selected as the best fit model and these parameters categorized as the most important variable in the study area. The ANN Model 1 provides better prediction than ANN Model 2 and ANN Model 3 in term of API prediction. Lastly, from the ANN Model 1, the new API equation model was developed. This research has verified that environmetric techniques and ANN analysis are highly practicable and effective for analysing large amount of complex data to gather vital knowledge about air quality, especially the behaviour characteristics of spesific air pollutants and air pollution patterns. Besides, this work showed the importance of historical data in sampling plan strategies to achieve desired research objectives, as well as to highlight the possibility of determining the optimum number of sampling parameters, which in turn will reduce costs and time of sampling. In other words, this knowledge can be applied by policy and decision makers as a tool for planning more effective air quality monitoring programs.
|
|---|