Utilisation critique de l'analyse en composantes principales et du cluster analysis pour la description d'échantillons d'invertébrés benthiques en eau douce
Principal component analysis and cluster analysis, applied to experimental data obtained by colonization of artificial substrates revealed clusters of ecological stations which, in the plane of the first 2 principal components, are localized successively on a non-linear structure correlated to a water quality gradient. Non linearity of interrelationships between species abundances produces distortions in the principal components, which prevent the complete description of a theoretical model, but the variability between samples is so large, that a technique using correlations between species is the only 1 which can mask the random noise. Principal component analysis is in fact a good tool for the classification of ecological stations in a reduced space. Biological and physico-chemical profiles of the ecological stations and their sequence on this non-linear structure, permit rediscovery of the water quality gradient and its effects on the biocenose. The results are compared with 2 classical biological indices.