Inter-comparison of source apportionment of PM10 using PMF and CMB in three sites nearby an industrial area in central Italy

Abstract

Receptor models (RMs), based on chemical composition of particulate matter (PM), such as Chemical Mass Balance (CMB) and Positive Matrix Factorization (PMF), represent useful tools for determining the impact of PM sources to air quality. This information is useful, especially in areas influenced by anthropogenic activities, to plan mitigation strategies for environmental management. Recent inter-comparison of source apportionment (SA) results showed that one of the difficulties in the comparison of estimated source contributions is the compatibility of the sources, i.e. the chemical profiles of factor/sources used in receptor models. This suggests that SA based on integration of several RMs could give more stable and reliable solutions with respect to a single model. The aim of this work was to perform inter-comparison of PMF (using PMF3.0 and PMF5.0 codes) and CMB outputs, focusing on both source chemical profiles and estimates of source contributions. The dataset included 347 daily PM10 samples collected in three sites in central Italy located near industrial emissions. Samples were chemically analysed for the concentrations of 21 chemical species (NH4+, Ca2+, Mg2+, Na+, K+, Mg2+, SO42−, NO3−, Cl−, Si, Al, Ti, V, Mn, Fe, Ni, Cu, Zn, Br, EC, and OC) used as input of RMs. The approach identified 9 factor/sources: marine, traffic, resuspended dust, biomass burning, secondary sulphate, secondary nitrate, crustal, coal combustion power plant and harbour-industrial. Results showed that the application of constraints in PMF5.0 improved interpretability of profiles and comparability of estimated source contributions with stoichiometric calculations. The inter-comparison of PMF and CMB gave significant differences for secondary nitrate, biomass burning, and harbour-industrial sources, due to non-compatibility of these source profiles that have local specificities. When these site-dependent specificities were taken into account, optimising the input source profiles of CMB, a significant improvement in the comparison of the estimated source contributions with PMF was obtained.