PM2.5 source apportionment during severe haze episodes in a Chinese megacity based on a 5-month period by using hourly species measurements: Explore how to better conduct PMF during haze episodes

To understand rapid physicochemical changes and to explore how to better conduct PMF for source apportionment during haze episodes, EPA Positive Matrix Factorization (PMF) 5.0, including an assessment of uncertainties, was performed based on hourly measurements of PM2.5-bound species during a heating period in 2017-2018 in the Chinese megacity of Tianjin. Five haze episodes were the focus of this study. PMF was conducted using two modes (whole-based mode for the entire sampling period and episode-based mode for each episode), and some indicators, including bootstrap (BS), displacement of factor elements (DISP), E/M (ratios of estimated and measured concentrations), were then used to evaluate PMF performance. The contributions of secondary ions and secondary organic carbon (SOC) for episodes I and III were greater than 50% (which were defined as SPE: episodes strongly influenced by secondary particles). The coal combustion contribution in episode II (defined as coal combustion episode: CCE) was greater than in the other episodes, and fireworks burning were extracted in episode IV (defined as fireworks burning episode: FBE). The PMF solutions were poor for the regional transport episode (RTE), during which episode the species showed weak variations (low coefficients of variation) and similar patterns (high correlations). During SPE, E/M were 0.71-1.00 for episode-based mode and 0.65-1.28 for whole-based mode, indicating good performance for both modes; while during CCE and FBE, E/M were 0.53-0.99 for episode-based mode and 0.10-1.06 for whole-based mode, showing that estimation of some important markers were poor for the whole-based mode. Sensitivity tests were then conducted to systematically investigate the influence of heavy-pollution types on PMF and showed that PMF was insensitive to contribution variations but was strongly sensitive to variations in source profiles. Overall, episode-based mode of PMF is better for episodes with strong variations of primary sources (CCE and FBE), whereas whole-based mode can be used for SPE and RTE. This work will help us to accurately assess hourly source variations.