Abstract. In January 2013, February 2014, December 2015 and December 2016 to 10 January 2017, 12 persistent heavy aerosol pollution episodes (HPEs) occurred in Beijing, which received special attention from the public. During the HPEs, the precise cause of PM2.5 explosive growth (mass concentration at least doubled in several hours to 10 h) is uncertain. Here, we analyzed and estimated relative contributions of boundary-layer meteorological factors to such growth, using ground and vertical meteorological data. Beijing HPEs are generally characterized by the transport stage (TS), whose aerosol pollution formation is primarily caused by pollutants transported from the south of Beijing, and the cumulative stage (CS), in which the cumulative explosive growth of PM2.5 mass is dominated by stable atmospheric stratification characteristics of southerly slight or calm winds, near-ground anomalous inversion, and moisture accumulation. During the CSs, observed southerly weak winds facilitate local pollutant accumulation by minimizing horizontal pollutant diffusion. Established by TSs, elevated PM2.5 levels scatter more solar radiation back to space to reduce near-ground temperature, which very likely causes anomalous inversion. This surface cooling by PM2.5 decreases near-ground saturation vapor pressure and increases relative humidity significantly; the inversion subsequently reduces vertical turbulent diffusion and boundary-layer height to trap pollutants and accumulate water vapor. Appreciable near-ground moisture accumulation (relative humidity> 80 %) would further enhance aerosol hygroscopic growth and accelerate liquid-phase and heterogeneous reactions, in which incompletely quantified chemical mechanisms need more investigation. The positive meteorological feedback noted on PM2.5 mass explains over 70 % of cumulative explosive growth.

Abstract. In order to reduce human efforts, there has been increasing interest in applying active learning for training text classifiers. This paper describes a straightforward active learning heuristic, representative sampling, which explores the clustering structure of 'uncertain' documents and identifies the representative samples to query the user opinions, for the purpose of speeding up the convergence of Support Vector Machine (SVM) classifiers. Compared with other active learning algorithms, the proposed representative sampling explicitly addresses the problem of selecting more than one unlabeled documents. In an empirical study we compared representative sampling both with random sampling and with SVM active learning. The results demonstrated that representative sampling offers excellent learning performance with fewer labeled documents and thus can reduce human efforts in text classification tasks.

No abstract

Abstract. Accompanied by unfavorable meteorological conditions with stable stratification in various haze regions of China, persistent heavy aerosol pollution episodes (HPEs) lasting more than 3 consecutive days frequently occur, particularly in winter. In the North China Plain (NCP), explosive growth of fine particulate matter smaller than 2.5 µm in diameter (PM2.5), which occurs during some HPES, is dominated by a two-way feedback mechanism between more unfavorable meteorological conditions and cumulative aerosol pollution. However, the existence of a two-way feedback mechanism such as this in other key haze regions in China is uncertain; these regions include the Guanzhong Plain (GZP), the Yangtze River Delta (YRD) region, the Two Lakes Basin (TLB; a large outflow basin connected to Hubei Province and Hunan Province), the Pearl River Delta (PRD) region, the Sichuan Basin (SB), and the Northeast China Plain (NeCP). In this study, using surface PM2.5 and radiation observations, radiosonde observations, and reanalysis data, we observed the existence of a two-way feedback mechanism in the six abovementioned regions. In the SB, this two-way feedback mechanism is weak due to the suppression of cloudy mid-upper layers. In the more polluted NCP, the GZP, and the NeCP, the feedback is more striking than that in the YRD, the TLB, and the PRD. In these regions, the feedback of worsened meteorological conditions on PM2.5 explains 60 %–70 % of the increase in PM2.5 during the cumulative stages (CSs). For each region, the low-level cooling bias becomes increasingly substantial with increasing aerosol pollution and a closer distance to the ground surface. With PM2.5 mass concentrations greater than 400 µg m−3, the near-ground bias exceeded −4 ∘C in Beijing and reached up to approximately −4 ∘C in Xi'an; this result was caused by accumulated aerosol mass to some extent. In addition to the increase in PM2.5 caused by the two-way feedback, these regions also suffer from the regional transport of pollutants, including inter-regional transport in the GZP, trans-regional transport from the NCP to the YRD and the TLB, and southwesterly transport in the NeCP.

No abstract