Application of the factor-analysis receptor model to simulated urban- and regional-scale data sets

Factor analysis has been used extensively to model the sources of ambient aerosol. In this study, simple urban- and regional-scale simulations showed that factor analysis may not always produce reliable results. The accuracy of apportionments of total mass depended on the rotation scheme used to transform the factors. Varimax-rotated solutions were generally independent of the degree of random error in the data, but were sensitive to collinearities in profiles, correlations of source strengths, and magnitudes of source strengths. Target-transformation factor analysis was more successful than varimax rotation when targets were similar to true profiles.In simple regional simulations, midwestern and northeastern sources were resolved qualitatively by varimax-rotation and quantitatively by target-transformation factor analysis. Signatures determined from principal-component analysis of ambient data in the northeastern U.S. and the Arctic resembled those determined independently.