Fairly processing rare and common species in multivariate analysis of ecological series. Application to macrobenthic communities from Algiers harbour.

Systematic sampling of communities gives rise to large contingency tables summing up possible changes in the assemblages' structure. Such tables are generally analysed by multivariate statistical methods, which are ill-suited for simultaneously analysing rare and common species (Field et al., 1982). In order to separately process species belonging to either of these categories, we propose a statistical method to select common species in a sequence of ecological surveys. It is based on a precise definition of rarity, and depends on a rarity parameter. In this work, this parameter will be optimised so that the sub-table of common species captures the essential features of the complete table as well as possible. In this way we analysed the spatio-temporal evolution of macrobenthic communities from the Algiers harbour to study the pollution influence during a year. The examination of the communities' structuring was done through Principal Components Analysis (PCA) of the species proportions table. Environmental variables were simultaneously sampled. We show that the data structure can be explained by about 25% of the total number of present species. Two environmental gradients were brought to the fore inside the harbour, the first one representing pollution, and the second one representing hydrological instabilities. Since rare species can also convey information, the complete table was also coded according to a generalised presence/absence index and submitted to Correspondence Analysis. The results were consistent with those of PCA, but they depended on more species, and highlighted the influence of sedimentology on the assemblages composition.