Improving apportionment of PM2.5 using multisite PMF by constraining G-values with a prioriinformation

Abstract

Rotational ambiguity in factor analyses prevents users from obtaining accurate source apportionment results. The rotational space in positive matrix factorization (PMF) can be reduced by constraining the solution with a prioriinformation such as source profiles. However, the only prior report on constraints using information on the source contributions was their use to ensure compatibility in the simultaneous analyses of PM2.5 and PM10 data. By combining data from three monitoring sites affected by a gear casting plant in Xi'an as an example, a methodology for improving the accuracy of PMF results by constraining source contributions using wind information was explored. Seven common factors derived from individual PMF analyses for each of the three sites (INDUS, URBAN, and RURAL) with different location characteristics, were then combined in a multisite PMF analysis. The factors were interpreted as nitrate with all site average contributions of 28.7%, sulfate (22.5%), coal combustion (19.3%), road traffic (12.8%), biomass burning (6.4%), soil (5.4%), and metallurgical industry (4.9%). Except for the INDUS site, contributions of metallurgical industry to the URBAN and RURAL sites were pulled down maximally to reduce the rotational space. The constrained solution substantially improved the results over the base run. The local and regional nature of the sources were identified by coefficient of divergence combined with Pearson correlation analysis, and further quantitatively estimated using Lenschow approach. On average, local sources contributed for 52.4% and 47.7% of the PM2.5 mass concentrations at the INDUS and URBAN site respectively. The metallurgical industry showed the highest local contributions while sulfate was primarily regional. For the multisite analysis where there are considerable point source emissions, this methodology highlights the role of local wind directions to inform constraints on the results and obtaining more reliable solutions.