MULTIVARIATE DATA-ANALYSIS APPLIED TO THE INVESTIGATION OF RIVER POLLUTION

A set of quantitative analytical data for the rivers Bormida and Tanar o in the Piedmont region (north-western Italy) has been processed by m ultivariate statistical techniques. The aim consists in evaluating the usefulness of these methods, compared with usual univariate technique s, in the investigation of surface water pollution phenomena. The expe rimental data consist of 19 chemical and bacteriological variables det ermined at 31 sampling sites. The following methods have been used for the treatment of the data: cluster analysis (unsupervised pattern rec ognition), principal component analysis, non-linear mapping, multivari ate feature selection. The treatment has been proved to be useful in d etermining the various factors of pollution together with their mutual importance and correlation in evaluating the general degree of pollut ion of rivers, and in rationalizing the data collection by eliminating variables with low information content.