Identifying and quantifying PM2.5 pollution episodes with a fusion method of moving window technique and constrained Positive Matrix Factorization

PM2.5 pollution episodes rapidly and significantly deteriorate the air quality and are a critical concern world-wide. This study developed a fusion method based on the moving window dataset technique and constrained Positive Matrix Factorization (PMF) to differentiate and characterize potential factors in a PM2.5 episode case assuming having one new contributor. The hourly PM2.5 compositions of elements, ions and carbonaceous components, were collected from September to December 2020 in Taipei, Taiwan. Constraint targets based on the bootstrap analysis result of a PMF model using a long-term input dataset were imposed on the modeling of each moving window to ensure similar features of the retrieved factors. The constituents of an additionally differentiated factor to the episode, which was identified as regional transport, were stable among each moving window that covered the occurrence of the episode as revealed by the profile matching index. The results showed that the largest contributor to the PM2.5 mass during the episode period of 12/12/2020 was regional transport (61%), whereas that of 12/13 was the regular pollution of industry/ammonium sulfate related (43%). According to our review of the literature, this study is the first to apply both the moving window technique and constrained PMF to characterize the episode. The findings provide valuable information that can be used to explore the causes of PM2.5 episodes and implement air pollution control strategies.