The effects of meteorological conditions and long-range transport on PM2.5 levels in Hanoi revealed from multi-site measurement using compact sensors and machine learning approach

• Published in: Journal of aerosol science 2021-02, Vol.152, p.105716, Article 105716
• Main Authors: Ly, Bich-Thuy, Matsumi, Yutaka, Vu, Tuan V., Sekiguchi, Kazuhiko, Nguyen, Thu-Thuy, Pham, Chau-Thuy, Nghiem, Trung-Dung, Ngo, Ich-Hung, Kurotsuchi, Yuta, Nguyen, Thu-Hien, Nakayama, Tomoki
• Format: Article
• Language: English
• Subjects: Aerosol; Long-range transport; Machine learning; Meteorology; PM2.5 sensor
• ISSN: 0021-8502; 1879-1964
• DOI: 10.1016/j.jaerosci.2020.105716

Hanoi, the capital of Vietnam, frequently experiences heavy air pollution episodes in the winter, causing health concerns for the 7.5 million people living there. Spatial-temporal variations in PM2.5 levels can provide useful information about the sources and transportation of PM2.5. However, the published spatial-temporal data in the area are limited. In this research, PM2.5 concentrations at two sites in Hanoi and a site in Thai Nguyen (60 km north of Hanoi) were observed from October 2017 to April 2018, using newly available low-cost sensors. Hourly concentrations of PM2.5 at the three sites were similar on average (57.5, 54.9, and 53.6 μg m−3) and clearly co-varied, suggesting remarkable large-scale effects. The contribution of long-range transport and meteorological factors on PM2.5 levels were investigated with a machine learning technique based on a random forest (RF) algorithm and concentration weight trajectory (CWT). The results showed that the contribution of long-range transport from the north and northeast to local PM2.5 levels was significant. Moreover, weather normalized PM2.5 concentrations and partial plots of meteorological factors on the levels of PM2.5 showed that meteorological conditions play a significant role in the formation of winter haze events. [Display omitted] •Observation of PM2.5 levels at three sites using sensors.•Moderate to good correlation factors among PM2.5 at sites showed regional effects.•Partial effects of meteorological and temporal factors on PM2.5 were determined by a machine learning approach.•Contribution of meteorological factors on haze revealed by weather normalized PM2.5•PM2.5 long-range transport investigated by CWT and partial trajectory correlation.