First, the heterogeneity in community composition must be tested with a DCA.
If the gradient length exceed 4.0, then the distribution is expected to be unimodal, which is required for a CCA. If the gradient is shorter than 3.0, a RDA (linear method) should be applied instead of CCA. Between 3.0 and 4.0, both CCA and RDA give similar results.
The unimodal methods, like CCA, must be performed with standardized data, unlike RDA which allows both standardized and non-standardized data.