—The prediction of each of air pollutants as dependent variable was investigated using lag-1(30 minutes before) values of air pollutants (nitrogen dioxide, NO2
, particulate matter 10um, PM10
, and ozone, O3
) and meteorological factors and temporal variables as independent variables by taking into account serial error correlations in the predicted concentration. Alternative variables selection based on independent component analysis (ICA) and principal component analysis (PCA) were used to obtain subsets of the predictor variables to be imputed into the linear model. The data was taken from five monitoring stations in Surabaya City, Indonesia with data period between March-April 2002. The regression with variables extracted from ICA was the worst model for all pollutants NO2
, and O3
as their residual errors were highest compared with other models. The prediction of one-step ahead 30-mins interval of each pollutant NO2
, and O3
was best obtained by employing original variables combination of air pollutants and meteorological factors. Besides the importance of pollutants interaction and meteorological aspects into the prediction, the addition spatial source such as wind direction from each monitoring station has significant contribution to the prediction as the emission sources are different for each station.
—Linear regression, principal component regression, independent component regression, air quality prediction, generalized least square.
The authors are with the Graduate School of International Development and Cooperation, Hiroshima University, Japan (e-mail: d115407@ Hiroshima-u.ac.jp, firstname.lastname@example.org, email@example.com).
Cite:Arie Dipareza Syafei, Akimasa Fujiwara, and Junyi Zhang, "Prediction Model of Air Pollutant Levels Using Linear Model with Component Analysis," International Journal of Environmental Science and Development vol. 6, no. 7, pp. 519-525, 2015.