Multivariate methods to predict ground level ozone during daytime, nighttime, and critical conversion time in urban areas

Abstract

Ground–level ozone (O3) is known to exhibit strong daily variations that lead to complexity of the pollutants’ analysis and predictions. This study aimed to introduce and explore the variations in O3 concentrations during daytime (DT), nighttime (NT), and critical conversion time (CCT) using multiple linear regression (MLR) and principal component regression (PCR) analyses. The original variables and principal component analysis (PCA) results were used as the input for MLR analysis. Hourly averages of six air pollutants and four meteorological parameters at Shah Alam during 1999–2009 were selected for this study. The monitoring records in 2010 were used to assess the developed models using several performance indicators. Results showed that the MLR model during DT exhibited optimal performance in terms of normalized absolute error, index of agreement, prediction accuracy, and coefficient of determination (R2) with values of 0.2762, 0.9211, 0.8581, and 0.7354, respectively. PCR during CCT also showed significantly higher performance than that during DT and NT. This result was evidenced by higher percentage of total variances, which could be explained by the selected variables in PCA during CCT.