Air Quality Modeling via PM2.5 Measurements

Environmental quality for the general population is dominated by air quality. Thus modeling of air quality is the first step toward any program for quality improvement. This paper describes the use of the ARIMA (Autoregressive Integrated Moving Average) time series modeling approach, illustrated by the tracking of the daily mean PM2.5 concentration in the north region of Singapore. A framework for ARIMA forecasting revised from the general Box-Jenkins procedure is first outlined; T-test and three information criteria, namely, AIC (Akaike Information Criterion), BIC (Bayesian Information Criterion), HIC (Hannon-Quinn Information Criterion) are employed in addition to analyses on the ACF (auto-correlation function) and PACF (partial autocorrelation function). With forecasting as the primary objective, the emphasis is on out-of-sample forecasting more than in-sample fitting. It is shown that for 30 such forecasts, one-step ahead MAPE (mean absolute percentage error) has been found to be as low as 8.0%. The satisfactory result shows the classical time series modeling approach to be a promising tool to model compound air pollutants such as PM2.5; it enables short-term forecasting of this air pollutant concentration for public information on air quality.