I guess your question is about the difference between Canonical Correspondence Analysis (CCA) and Canonical Redundancy Analysis (RDA), right? Canonical Correspondence analysis is a type of Canonical Analysis.
Choosing between CCA and CCA to biodiversity studies should be based on the type of response you expect from the biodiversity matrix. If you expected linear responses, use RDA. If you expect unimodal responses you should chose CCA. Of course, you can transform a non-linear data and use the RDA, which usually has few problems than CCA.
I think you are trying to distinguish Canonical Correlation Analysis and Canonical Correspondence Analysis. They are similar, in that they both search for a multivariate relationship between two datasets (like an environmental dataset and a species abundance dataset). Canonical Correlation Analysis assumes a linear relationship, which is often not found in nature. Population ecologists generally use Canonical Correspondence Analysis, which assumes a unimodal response curve.
Are you asking about the difference between Correspondence Analysis (CA) and Canonical Correspondence Analysis (CCA)? If so, the latter is the constrained form (ie, using explanatory variables to explain a set of response variables) of the latter (ie, no explanatory variables, in which patterns are revealed by the 'response' variables only) on a Chi-square space (ie, your data come from sufficiently long ecological gradients and thus exhibit a unimodal response, think about niche theory).
In either case both are useful, but the application of one or the other will depend on your research questions, hypothesis and ultimately your data. For instance, if you have no explanatory variables, you won't be able to use CCA. Hope this helps.
You can see Multivariate Statistatical Methods A Primer (Manly). It is very easy and clear. The another one is Multivariate Analysis in Community Ecology Gauch.
Remember that a Canonical Correspondence analysis is a type of Canonical Analysis,
Informative responses! Please how do i deal with data that is not normally distributed, even after transformation, when i need to carry out canonical correspondence analysis? Also, must there be significant correlation before CCA can be employed?