Evaluating the influence of constant source profile presumption on PMF analysis of PM2.5 by comparing long- and short-term hourly observation-based modeling

Hourly PM₂.₅ speciation data have been widely used as an input of positive matrix factorization (PMF) model to apportion PM₂.₅ components to specific source-related factors. However, the influence of constant source profile presumption during the observation period is less investigated. In the current work, hourly concentrations of PM₂.₅ water-soluble inorganic ions, bulk organic and elemental carbon, and elements were obtained at an urban site in Nanjing, China from 2017 to 2020. PMF analysis based on observation data during specific pollution (firework combustion, sandstorm, and winter haze) and emission-reduction (COVID-19 pandemic) periods was compared with that using the whole 4-year data set (PMFwₕₒₗₑ). Due to the lack of data variability, event-based PMF solutions did not separate secondary sulfate and nitrate. But they showed better performance in simulating average concentrations and temporal variations of input species, particularly for primary source markers, than the PMFwₕₒₗₑ solution. After removing event data, PMF modeling was conducted for individual months (PMFₘₒₙₜₕ) and the 4-year period (PMF₄₋yₑₐᵣ), respectively. PMFₘₒₙₜₕ solutions reflected varied source profiles and contributions and reproduced monthly variations of input species better than the PMF₄₋yₑₐᵣ solution, but failed to capture seasonal patterns of secondary salts. Additionally, four winter pollution days were selected for hour-by-hour PMF simulations, and three sample sizes (500, 1000, and 2000) were tested using a moving window method. The results showed that using short-term observation data performed better in reflecting immediate changes in primary sources, which will benefit future air quality control when primary PM emissions begin to increase.