EXTRACTING INFORMATION FROM COMPLEX CHROMATOGRAPHIC FINGERPRINTS FOR EVALUATION OF ORGANIC AIR-POLLUTION

The gas chromatographic profiles of 47 samples of airborne particulate matter, obtained by flame ionization detection, were synchronized and quantified into variables by an automatic procedure, avoiding peak re cognition and identification. Pattern recognition methods of principal components and clustering were used to extract information from the v ariables and seasonal similarities were established. Multivariate regr ession with the partial least squares method found predictive relation ships between the profiles and other variables determined by different methods: extractable organic material, carbon preference index of the n-alkane homologous series (computed from gas chromatographic-mass sp ectrometric determinations) and mutagenicity. The predictive power was between 68% and 81%, and so the usefulness of the extracted informati on was verified; the profile obtained by flame ionization detection co ntains information related to the quality of air. The procedure is als o suitable for different kinds of complex matrices, provided that sync hronization is realized: it gives a first general knowledge of sample characteristics which may be used to address further analysis.