Influence of boundary layer structure on air quality in Beijing: Long-term analysis based on self-organizing maps

Abstract. Self-organizing maps (SOMs; a feather-extracting technique based on an unsupervised machine learning algorithm) are used to classify the atmospheric boundary layer (ABL) types over Beijing by detecting topological relationships among the 4-yr (2013–2016) radiosonde profiles. The resulting ABL types are then examined in relation to air quality, including surface pollutant concentrations and columnar aerosol properties, to understand the regulating effects of different ABL structures on Beijing's air quality. The SOM provides nine ABL types (i.e., SOM nodes), and each type is characterized by distinct dynamic and thermodynamic conditions. On average, SO2, NO2, CO, PM10 and PM2.5 increase 120–220 % from a near neutral (i.e., node 1) to strong stable condition (i.e., node 9). The ABL controls on diurnal cycles of pollutants are as follows: (1) elevated inversion enhances the afternoon baseline; and (2) surface inversion improves the evening increment. Comparing the CO / SO2 ratios for the different ABL types demonstrates that the local contribution increases with enhanced static stability near the ground, and it is the stable ABL stratification rather than weak surface wind that confines the regional contribution. Due to regional transport, node 3 (dominated by elevated inversion with high relative humidity) corresponds to the most severe columnar aerosol pollution, characterized by the highest optical depth (1.22) and volume concentration (0.30 μm3/μm2). The larger aerosol radiative forcing (ARF) within the atmosphere (> 60 W/m2) in nodes 3, 6 and 9 is likely to strengthen the atmospheric stability and thus induce a positive feedback loop for causing high surface pollution. Analysis of the typical pollution period suggests that the ABL types are the primary drivers of day-to-day variations in Beijing's air quality. Assuming a fixed relationship between ABL type and PM2.5 loading for different years, the relative (absolute) contribution of the ABL anomaly to elevated PM2.5 levels are estimated to be 65.8 % (46.2 μg/m3) during January 2013, 46.7 % (20.2 μg/m,sup>3) during December 2015, and 94.6 % (35.3 μg/m3) during December 2016.