The analysis of pollutant profiles by means of compositional analysis. An application to PCDD/F data

Compositional analysis consists of a group of techniques used to manipulate closed (=compositional) data, i.e. multivariate data summing to a fixed quantity (proportions, percentages). It is based on the analysis of the relations among variables and the use of logarithmic transformations. It has been claimed that this group of techniques should be used to analyse profiles of pollutant sources because profiles themselves are proportions. We show in this paper that, for the exploratory analysis of these data, a good strategy is to combine the analysis done with and without transformation because they give different and complementary insights on the structure of the data. We discuss in particular the study of processes such as the mixing of pollutants produced by different sources and the exponential decay of concentrations with distance to the source found in many studies. The clr transformation is also appropriate for the study of the variables having small proportions, which remain concealed by the abundant variables when analysed without transformation. We present simulations to illustrate these ideas and we also apply these techniques to two data sets of PCDD/F content in moss tissues.