How do you represent community data as a predictor?

[wcornwell]

This is for #6 and/or #7. I'm sure there is an established technique for this, but just not sure what it is. Any thoughts @jeffpowell2 ?

[jeffpowell2]

We've done this before using the function WAPLS in the package rioja, but it was a while ago so I need to revisit that code.

From ter Braak and Schaffers (2004, http://onlinelibrary.wiley.com/doi/10.1890/03-0021/abstract):

Weighted averaging calibration (Table 1) predicts an environmental variable from community data by averaging indicator values of the species present at a site (e.g., Persson 1981, Jongman et al. 1995). When the indicator values (the species scores of this method) for a particular environmental variable are unknown, they can be estimated from training data by weighted averaging regression (ter Braak and van Dam 1989, Birks et al. 1990, Fritz et al. 1991). The PLS extension of weighted averaging regression and calibration (WA-PLS; ter Braak and Juggins 1993, ter Braak et al. 1993) uses ideas from PLS to estimate the indicator values. WA-PLS is popular in palaeoecology for reconstructing palaeoenvironments from fossil assemblages (Birks 1998). WA-PLS is called correspondence analysis partial least squares by Frisvad and Norsker (1996). The community data take the role of response variables in CCA-PLS and the role of predictor variables in WA-PLS (Table 1).

[wcornwell]

I have asked David Warton whose suggestion was very similar to @jeffpowell2 ... also could try (for completeness) some of the BEF type metrics (richness, phylogenetic diversity, etc).

[marissalee]

....went with WA-PLS in the rioja package