Multivariate analysis of monsoon seasonal variation and prediction of particulate matter episode using regression and hybrid models

Prediction of particulate matter (PM₁₀) episode in advance enables for better preparation to avert and reduce the impact of air pollution ahead of time. This is possible with proper understanding of air pollutants and the parameters that influence its pattern. Hence, this study analysed daily average PM₁₀, temperature (T), humidity (H), wind speed and wind direction data for 5 years (2006–2010), from two industrial air quality monitoring stations. These data were used to evaluate the impact of meteorological parameters and PM₁₀ in two peculiar seasons: south-west monsoon and north-east monsoon seasons, using principal component analysis (PCA). Subsequently, lognormal regression (LR), multiple linear regression (MLR) and principal component regression (PCR) methods were used to forecast next-day average PM₁₀ concentration level. The PCA result (seasonal variability) showed that peculiar relationship exists between PM₁₀ pollutants and meteorological parameters. For the prediction models, the three methods gave significant results in terms of performance indicators. However, PCR had better predictability, having a higher coefficient of determination (R²) and better performance indicator results than LR and MLR methods. The outcomes of this study signify that PCR models can be effectively used as a suitable format in predicting next-day average PM₁₀ concentration levels.