Modelling particulate matter (PM10) concentration in industrialized area: A comparative study of linear and nonlinear algorithms

Abstract

Particulate matter is a critical air pollutant in Malaysia as it is the utmost dominant pollutant, especially in industrial and urban areas. The development of a robust model for PM10 concentration forecasting provides invaluable information for local authorities to take precautionary measures and implement significant actions to improve air pollution status. This study aims to develop and assess the linear (Multiple Linear Regression, MLR) and nonlinear (Multilayer Perceptron, MLP) models forecasting capability in industrial area of Pasir Gudang, Johor. Daily observations of PM10 concentration, meteorological factors (wind speed, ambient temperature and relative humidity) and gaseous pollutants (SO2, NO2 and CO) from the year 2007-2014 were used in this study. Results showed that MLP model was able to explain 68.7% (R2 = 0.687) variance in the data compared to MLR model with 52.7% (R2 = 0.527). Overall, the MLP model able to increase the accuracy of forecasting by 29.9% and reducing the error by 69.3% with respect to MLR model. Thus, it is proven that nonlinear model has high ability in virtually representing the complexity and nonlinearity of PM10 in the atmosphere without any prior assumptions, unlike the linear model. © 2006-2018 Asian Research Publishing Network (ARPN).