Quantile regression for analysing PM10 concentrations in Petaling Jaya

Abstract

Particulate matter with diameter less than 10µm (PM10) data usually exhibit different variations as they include normal days and pollution days. This paper applied quantile regression (QR) technique to inspect the changing relationship between predictor variables and PM10 concentrations at Petaling Jaya monitoring station in the year 2014 over different PM10 distributions. For comparative purpose, multiple linear regression (MLR) using ordinary least squares (OLS) estimation approach was also performed. The QR analysis results showed that the interrelationship between predictor variables and PM10 was not consistent across the PM10 quantile distributions and hence, proved discordancy with MLR estimates. The lagged PM10 concentration was the only important factor throughout the quantile distributions of PM10. It was found that the effects of lagged PM10, temperature, carbon monoxide (CO) increased from low to high quantile distributions, while the effects of lagged humidity, east-west wind component, wind speed and nitrogen monoxide (NO) showed the otherwise patterns. The lagged NO associated significantly with PM10 at low quantiles, whereas the lagged temperature and CO associated significantly at high quantiles only. Lagged humidity, east-west wind component and wind speed correlated significantly and negatively with PM10 at low and middle quantiles. Ozone (O3), however, had effect of changing nature from positive association at low PM10 distributions to negative association at high levels. Thus, QR is helpful to provide a more complete description of predictor variable effects on PM10 at different distributions, and may assist in PM10 management especially during haze periods.