Noise Reduction By Eigenvector Ordinations

Abstract

Field data on the species content of plant and animal communities are noisy. Variation in community samples partly reflects interesting variation in environmental and historical factors, and partly reflects random fluctuations in species abundances. Routinely community data are analyzed by eigenvector ordination techniques, such as principal components analysis, reciprocal averaging, and detrended correspondence analysis. It is shown here with simulated community data that ordi- nation selectively recovers patterns affecting several species simultaneously in early ordination axes, while selectively deferring noise to late axes. Eigenvector ordinations thus appear to be effective for reducing noise. This result helps to explain the observation that ordinations of field data are frequently useful even when the percentage of variance accounted for by the first few ordination axes is small. A related conclusion is that rounding of the abundance values of community data sets has little effect on results from ordination, and consequently fairly crude field data are entirely adequate for ordination purposes.