Influence of the sampling period and time resolution on the PM source apportionment: Study based on the high time-resolution data and long-term daily data

When planning short-term and long-term measurement campaigns of particulate matter (PM), parameters such as sampling period, time resolution, sampling number, etc. are vital. To study their influence and to provide suggestion for the sampling plan of PM source apportionment (SA), ambient and synthetic speciated datasets (including a high time-resolution dataset and a long-term daily dataset) were studied. First, aiming at studying the sampling period required to generate representative and reliable results for SA, high time-resolution ambient samples were collected by online instruments in a megacity in China. Datasets with different sampling periods (four months, two months, one month, two weeks and one week) were modeled by the Positive Matrix Factorization (PMF). Compared with four month results, AAEs (percent absolute errors between true and estimated contributions) ranged from 11.2 to 27.2% (two months), 19.8-44.5% (one month), 21.0-45.9% (two weeks) and 23.9-44.6% (one week), indicating that divergence increased with decreasing sampling periods. To systematically evaluate this problem and investigate if the increasing time resolutions in a short period could enhance the modeling performance, synthetic datasets were constructed. Results revealed that a sufficient sampling period is required to ensure stable results; without sufficient sampling period, the contributions cannot be reliably estimated, even if the number of samples is large. Then, to explore the influence of variability absences, long-term daily datasets with various variability absences were apportioned and compared. The summed AAEs were 102.2% (no winter), 73.6% (no weekend), 138.7% (no weekday) and 165.6% (no autumn, winter or weekends). This general increase of AAEs can indicate that uncertainty enhanced with the increase in variability absences. When planning short-term measurement campaigns, except for number of samples, sampling period that involves sufficient source cycles has significant implications; when planning long-term sampling, more intensive sampling would increase the model performance.