Multivariate analysis for the reconstructions in palaeobiogeography

A large number of dates (14С, AMS) have been accumulated in the Late Pleistocene palaeontological databases of Europe. For example, the PALEOFAUNA database alone (Markova et al., 1995; 2008) includes about 3300 dates obtained from 300 mammalian species from different localities. Traditional methods appeared unworkable when analyzing such data sets. Applying new methods, we have obtained new information about spatial-temporal changes in mammalian assemblages during the Late Pleistocene. As a part of the NWO-RFBS project, nonparametric multidimensional scaling (MDS) and cluster analysis have been successfully used to reconstruct Late Weischelian mammal assemblages and analyse their structure and their ranges within Europe. MDS is a special case of ordination (Kruskal, 1964). Data analysis included several steps: testing of the data distribution, choice of metrics, computation of dissimilarities between localities, estimation of the number of the virtual ‘factors’ (MDS axes) and their interpretation. A good non-linear correlation between MDS axes and geographical coordinates of the localities is revealed in this analysis. Hypothetically, this result reflects the existence of an energetically powerful ‘mechanism’ of ‘spatial order’ during the last glaciation.