Some Comments on Niche Analysis in Canonical Space

Abstract

Discriminant functions have been used to identify axes of niche separation, and dispersion of locations on these axes have, in turn, been interpreted as representing niche width and specialization. We discuss the influence of sampling scheme and choice of niche measures on the results of an analysis conducted in canonical space. Specifically, habitat variables measured at random locations should provide a more representative measure of habitat available than would recording habitat only where animals are observed. Species presence information alone tends to bias the mean habitat toward the one where the most common species is found. If niche specialization is to be measured by distance from the overall mean habitat, each sampling station should receive equal weighting. The most straightforward measure of habitat breadth seems to be the variance or standard deviation of canonical scores. The mean squared distance from the species centroid can then be used as a comparable multidimensional breadth measure. Finally, we suggest that niche overlap be measured via a joint probability density function rather than by area of overlap of concentration ellipses.