Estimating uncertainties of source contributions to PM2.5 using moving window evolving dispersion normalized PMF

Abstract

Conventional factor analyses can present problems in cases with changing numbers of sources and/or time-dependent source compositions. There is also lack of a reliable method to estimate uncertainties in the source contributions derived by positive matrix factorization (PMF). Applying a moving window evolving PMF to hourly PM2.5 composition dataset from a field campaign in Tianjin China that included the Spring and Lantern Festivals and the start of COVID-19 pandemic has substantially improved the apportionment compared to a conventional analysis using the entire data. Festival-related sources (e.g., fireworks and residential burning here) have been clearly identified and estimated during both the Spring and Lantern Festivals. During this period, the sources changed because the time period overlaps with the outbreak of COVID-19 and related reductions in activity during the lockdown that began on Lunar New Year. Multiple PMF runs providing source contribution estimates made it possible to estimate the uncertainties in these values. Our results show that winds-dependent sources like dust and distant point sources have larger uncertainties than the other sources. Compared with conventional PMF analyses, the current method may better reflect the actual emissions as well as being able to estimate uncertainties. Thus, this approach appears to be an improvement if the appropriate data are available.