Multivariate statistical analysis of ecological communities (review)
V.K. Shitikov, Т.D. Zinchenko
Section: Theoretical problems of ecology
The modern approach to the study of communities’ ecology involves the integration and joint processing of large
arrays of observations. This usually involves a variety of indicators (population, phenotypic, genetic, environmental,
chemical, landscape and geographical) which are characterized by significant temporal and spatial variability. The purpose
of the analysis is to identify significant statistical relationships of the taxonomic structure with the characteristics
of biotopes and environmental factors. It is based on multivariate methods that allow optimal projection of data with a
large number of variables into low-dimensional spaces.
The article focuses on the evolution of algorithms for multivariate analysis, starting with the classic unconstrained
ordination based on principal components (PCA) up to modern integrated symmetric methods used in omics technologies.
A class of algorithms, such as metric (PCoA) and non-metric (NMDS) multidimensional scaling, based on the calculation of distance matrices, is distinguished and the advantages and disadvantages of their use are considered. The dependence
of the results of redundancy analysis (RDA) and canonical correspondence analysis (ССA) on the distribution law of the
empirical data is discussed and recommendations for their preliminary transformation are given.
It is shown the role of such symmetric methods as the two-block algorithm of partial least squares (2B-PLS) and
the co-inertia analysis (CIA), which allow to establish by decomposition on axes of multidimensional covariations what
species from different complexes of observations are most associated among themselves. Procrustean analysis (PCIA)
can be widely used to identify changes in the species composition of the study region before and after some event (e.g.,
anthropogenic impact). Generalized Procrustean algorithms and canonical analysis (GPA, RGCCA, DIABLO) allow you
to work with a large number of tables and explore the dynamics of community structure for several sequential periods of
time or to form a consensus configuration by the best way.
References to numerous examples of the use of ordination methods in domestic and foreign literature are given. The
main prospects and directions of development of multidimensional methods in relation to the ecology of communities
are shown.