AN INTERCOMPARISON OF PIXE AND XRF FOR STATISTICALLY RESOLVING AEROSOL COMPONENTS IN LENINGRAD AND ENVIRONS

PIXE and X-ray fluorescence methods for elemental analysis of aerosol samples have been intercompared from the standpoint of suitability for resolving principal components using multivariate statistical procedures. Agreement between the two analytical methods is excellent. However, important differences can result from differences in methods of aerosol sampling chosen for the two analytical methods, especially by sampling in separated coarse and fine particle size fractions by 2-stage streaker for PIXE, compared with single filter sampling for XRF without particle size selectivity. The PIXE streaker approach, requiring greater analysis effort in the laboratory but simpler sample collection procedures in the field owing to its automatic operation, provided more detailed statistical resolution of aerosol principal components than was attained by the XRF filter approach. The physical significance of these components is supported by independent particle number concentrations counted in 11 size ranges and resolution of principal components and calculation of their particle size distributions. Rather than emphasizing absolute accuracy of concentrations in air, our intercomparison depends on interelemental correlations of samples collected in time series and the precision of relative elemental determinations in relation to the sample-to-sample variability, the degree to which the elements are characteristic of the components and are determined with precision sufficient for their resolution, and the precision of component resolution in relation to the numbers of these elements and of the numbers of samples analyzed. A linear relation between characteristic X-rays detected and abundance of an element is required, but absolute calibration of X-rays with respect to amounts of an element is not. Atmospheric concentrations were measured in 56 consecutive 3-h samples collected for one week at two sites, in urban Leningrad and at nonurban Gorkovskaya 65 km north. Two-stage circular streaker samples and manually changed filters were analyzed by PIXE and XRF, respectively. Then a combination of factor and multiple linear regression analysis was carried out using a microcomputer and Statgraphics software for an IBM PC, assuming a linear receptor model in which data variance is accounted for by mixing components of fixed and distinctive elemental compositions. © 1990.