Nuclear Instruments and Methods in Physics Research B3 (1984) 483-488 North-Holland, Amsterdam 483 LONG RANGE AEROSOL TRANSPORT IN SOUTHERN SWEDEN: AN EXAMPLE OF MULTIVARIATE STATISTICAL EVALUATION METHODOLOGY Hans-Christen HANSSON, Bengt G. MARTINSSON and Hans 0. LANNEFORS Dept. of Nuclear Physics, Lund Institute of Technology, Siilvegatan 14, S-223 62 Lund, Sweden The utilization of multivariate statistical techniques is discussed with emphasis on the rather new method SIMCA, when applied to multielemental data. The procedures of scaling and normalizing are described. The data base used is from a project studying long range aerosol transport to southern Sweden. SIMCA reveals low variability in fine mode elemental composition in southerly air masses being clearly different from the elemental compositions found in northerly air masses. zyxwvutsrqponmlkjihgfedcbaZYXWVUT 1. Introduction Multivariate statistical techniques have been applied to air pollution data with some success during the last decade. The most widely used methods are principal component and factor analysis while in some cases cluster analysis has been used. These techniques have mostly been used in an explorative hypothesis generat- ing way in order to find correlating parameters in the data base studied, e.g. source patterns [1,2]. Factor and principal component analysis combined with other statistical techniques have been used in at- tempts to achieve more quantitative results. Henry and Hidy [3] have used multiple linear regression between a dependent variable and the principal components ob- tained by analysis of a set of other variables. Another interesting method is target transformation factor analy- sis [4]. This method is aimed at quantifying pollution sources. The factor rotation usually done after the fac- tor analysis is steered in the sense that the resulting factors should, as closely as possible, resemble any of a set of test vectors. These preliminary factors have a source characteristic elemental composition that could be altered during the iterative rotation procedure to achieve the best possible agreement. One problem with this method is that the solutions obtained are not unique. This work utilizes a recently developed multivariate statistical program package named SIMCA [5], which combines a pattern recognition technique and principal component analysis. The aim is to demonstrate how this multivariate technique can be applied to multielemental data, i.e. a typical PIXE data set. Important procedures like normalization with respect to meteorological in- fluence and scaling are discussed. A more detailed evaluation of the data set is presented in ref. 6. 0168-583X/84/$03.00 0 Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division) 2. Experimental 2. I. Sampling and analysis The sampling sites Ri%-vik and Falsterbo (see fig. 1) were chosen to represent locally undisturbed, westerly and southerly aerosol transport to southern Sweden. In zyxwvutsrq I SAMPLING SITE *MAJOR CITY & DSTEEL INDUSTRY 1 Fig. 1. Map showing the location of the sampling sites (Fal- sterbo, R&v& and SjWngen), some major cities and the Swedish steel industry locations. (The area contains several industries.) IV C. AEROSOL APPLICATIONS