Source apportionment using receptor model based on aerosol mass spectra and 1 h resolution chemical dataset in Tianjin, China

Source apportionment studies have been performed on online receptor datasets in recent years (called online source apportionment), including mass spectra and online chemical dataset. Single particle aerosol mass spectrometry (SPAMS), an important online technique, has the ability to analyze mass spectrum (MS) and particular size information of a single particle in real time. Clustering methods have been widely applied to MS dataset to investigate the sources of particles, although the receptor models are the common tools to probe the particle sources based on the receptor dataset. This work developed a new method (SPANS-FtM) that employed the receptor model (RM) on an MS dataset from SPAMS to identify particle sources. Particles were measured by SPAMS from July 14 to August 15, 2015, at an urban site in Tianjin, China. Multilinear Engine-2 (ME2) and adaptive resonance theory-based neural networks-2a (ART-2a) were separately used to analyze the single particle MS dataset. This work also evaluated the performance of SPAMS-RM method. Concentrations of chemical components of PM2.5 (particulate matter with an aerodynamic diameter of less than 2.5 mu m) and gaseous pollutants were measured by independent online instruments (1 h resolution). Source apportionment was separately conducted using two receptor models, Positive Matrix Factorization (PMF) and ME2, based on the 1 h resolution chemical dataset. This method was called online chemical source apportionment (OCSA-RM). ART-2a obtained 19 clusters that merged into five major classes: carbon species, rich-K, sea salt, crustal dust, and industrial metals. SPAMS-RM identified eight sources by interpreting the MS characteristic of factors and investigating the relationship of the temporal trends of factor contributions, chemical species, gaseous pollutants, and particle clusters. OCSA-ME2 and OCSA-PMF both identified seven factors. Source apportionment results between SPAMS-RM and OCSA-ME2/PMF were compared. Each method identified coal combustion, biomass burning, sea salt, nitrate source, sulfate source, vehicle emission, and crustal dust. The SPAMS-RM results showed that nitrate source was the most significant contributor (34%) to the PM followed by sulfate source (17%), coal combustion (14%), crustal dust (11%), vehicle emission (10%), biomass burning-OCEC (7%), and industrial activities & sea salt (4%). Some differences between SPAMS-RM and OCSA-ME2/PMF results existed and might be due to chemical analysis methods and sampling methods. ME2 was used for the first time to identify the PM sources based on the MS dataset from SPAMS and demonstrated its capability when coupled with MS dataset from SPAMS to apportion the source of PM.