LAUSR

Many source apportionment models have been applied to identify pollution sources, and differences often exist in the diagnostic results. The reasons causing these differences have not been fully elucidated. In this study, three receptor models, principal component analysis-multiple linear regression (PCA-MLR), positive matrix factorization (PMF), and factor analysis-nonnegative constraints (FA-NNC), were compared and applied for the analysis of 16 EPA priority polycyclic aromatic hydrocarbons (PAHs) adsorbed in street dust samples from Harbin City (China). The differences in the results were caused by different calculation approaches, including matrix decomposition, variable grouping extraction, and nonnegative constraints, especially between PCA-MLR and the other two models. PCA-MLR has no nonnegative constraints, making PCA-MLR less similar to the real world than the other two. Both PMF and FA-NNC have a nonnegative constraint process, which may be the main reason why their results were much more similar to each other than to those of PCA-MLR. PCA-MLR distinguishes variables into several groups that have the greatest variances from each other, whereas the other two methods find similarities among variables and extract them. In the case study of Harbin City, the contributions of mobile and industrial sources ranged from 47 to 69%, and the contributions of coal and other sources ranged from 30 to 52%. The recognized types of pollution sources were generally equivalent, but the proportional contributions were different. PCA-MLR performed best in calculating contributions, whereas PMF and FA-NNC were better in terms of source diagnosis.