The question of optimal transformation of species abundances in ordination has not yet been fully addressed. Techniques such as DCA and CCA appear to work well for raw data values (e.g. percent cover, biomass, basal area, etc.) as well as for logarithmic transformations, square root transformation, or transformation into a traditional cover-abundance scale. DCA and CCA also work well on presence/absence data. I usually use a square root transformation if I wish to dampen the effects of dominant species.
One commonly used transformation for species is log(x+k), where x is the species abundance and k is a constant, usually 1. The constant is added to avoid calculation of the logarithm of zero, which is undefined. I generally avoid this transformation because it will give different results if different units are used. For example, the log(x+k) transformation will give different answers if you record biomass in grams, kilograms, or pounds. Most other transformations do not suffer from this problem.
For CA, DCA, and CCA, it is not possible to have negative numbers for species abundances. This puts some constraints on the transformations used.
For linear methods (PCA, RDA) it is possible to have negative numbers. Indeed, if you standardize your species abundances (by subtracting the mean and dividing by the standard deviation, see Basic Statistics) you automatically have a PCA/RDA on the correlation matrix (see PCA and Similarity, Difference, and Distance). However, the easiest way to implement PCA/RDA on the correlation matrix in CANOCO is to select the option "Center and Standardize by Species".
The transformation of Environmental variables have different issues associated with them.