NEnvironmental variables were transformed (log10 + 1)
to normality when necessary. Variables that could not be
transformed, or were highly correlated (ρs
> 0.9), were
removed from subsequent analyses. Environmental data
were ordinated using a correlation-based principal components analysis (PCA). To find a subset of environmental
variables that best accounts for the faunal patterns a nonparametric analogue of canonical analysis, based on rank
similarities, was used called BIOENV (Clarke & Ainsworth 1993). Using the Spearman rank coefficient, pairwise correlations were performed between each of the
environmental variables and the fauna univariate measures