Influence of Noisy Environmental Data on Canonical Correspondence Analysis

Canonical Correspondence Analysis (CCA) is an increasingly popular method for multivariate analysis of ecological community data. It is, however, one of the most potentially misleading multivariate methods for community analysis. Inclusion of noisy or irrelevant environmental variables can distort the representation of gradients in community structure. These hazards are illustrated with simulated community data sets having a known, simple, underlying structure, then introducing different kinds and degrees of noise into the environmental data. Because of its sensitivity to even a modest amount of noise in the environmental data, CCA with site scores as linear combinations of environmental variables is inappropriate when the objective is to describe community structure. These problems can be avoided by using traditional indirect ordination methods, where pure community structure is expressed, without any constraint imposed by the environmental variables. CCA can be appropriate, however, when the objective is to describe how species respond to particular sets of observed environmental variables.