Receptor models application to multi-year ambient PM 10 measurements in an industrialized ceramic area: Comparison of source apportionment results

Abstract

Ambient PM 10 data collected in one of the largest industrialized ceramic areas of Europe were used to study similarities and differences in the source apportionment results from three widespread receptor models: chemical mass balance (CMB), positive matrix factorization (PMF) and principal component analysis (PCA). Particulate emissions were collected from a variety of sources including soil dust and different mixed raw materials used for the manufacture of ceramic tiles in the area. The chemical profiles of these emission sources are presented in this work. The analysis of the PMF scaled residuals was used as a diagnostic tool for adjusting species uncertainties and to assess the PMF model fit by comparison with the robust CMB results. The Q robust value, the degree of correlation between the predicted and measured species concentrations, the sample-by-sample correlation of the PMF source contributions compared with the CMB improved after the new error structure was used within the PMF model. The robustness of the CMB analysis used for the comparison with the PMF analysis was inspected by means of the CMB performances parameters as well as by comparing the results with a previous CMB analysis performed on the same database but with different speciated source profiles. Moreover, the results showed that PMF and PCA models were not able to distinguish between the two most important sources of crustal material in the selected area (one natural and one anthropogenic). With the CMB model a contribution from both sources was calculated without observing collinearity between the profiles. However, high correlation was found by adding the two crustal contributions from CMB and comparing the results with the single crustal factor from PCA and PMF. Low correlation was observed between the contribution values of the vehicular source for each model pairs. The lack of a local vehicular experimental profile for the CMB analysis and the non-specific chemical speciation performed for the ambient organic matter explained the low observed correlation.