The effect of the averaging period for PMF analysis of aerosol mass spectrometer measurements during offline applications

Abstract

Offline aerosol mass spectrometer (AMS) measurements can provide valuable information about ambient organic aerosols in areas and periods in which online AMS measurements are not available. However, these offline measurements have a low temporal resolution, as they are based on filter samples usually collected over 24 h. In this study, we examine whether and how this low time resolution affects source apportionment results. We used a five-month period (November 2016-March 2017) of online measurements in Athens, Greece, and performed positive matrix factorization (PMF) analysis to both the original dataset, which consists of 30 min measurements, and to time averages from 1 up to 24 h. The 30 min results indicated that five factors were able to represent the ambient organic aerosol (OA): a biomass burning organic aerosol factor (BBOA), which contributed 16 % of the total OA; hydrocarbon-like OA (HOA) (29 %); cooking OA (COA) (20 %); more-oxygenated OA (MO-OOA) (18 %); and less-oxygenated OA (LO-OOA) (17 %). Use of the daily averages resulted in estimated average contributions that were within 8 % of the total OA compared with the high-resolution analysis for the five-month period. The most important difference was for the BBOA contribution, which was overestimated (25 % for low resolution versus 17 % for high resolution) when daily averages were used. The estimated secondary OA varied from 35 % to 28 % when the averaging interval varied between 30 min and 24 h. The high-resolution results are expected to be more accurate, both because they are based on much larger datasets and because they are based on additional information about the temporal source variability. The error for the low-resolution analysis was much higher for individual days, and its results for high-concentration days in particular are quite uncertain. The low-resolution analysis introduces errors in the determined AMS profiles for the BBOA and LO-OOA factors but determines the rest relatively accurately (theta angle around 10° or less).