I am working with abiotic soil data such as bulk density, moisture levels and soil chemistry as response data (some quantitative some as percentages) and a mix of abiotic and biotic data as environmental data. To compare different sites I want to use multivariate statistics. For this PCA is an accepted method in this field, but with PCA I get an arch/horse shoe effect in my ordination plots.
I have been pointed to nMDS Non-metric multidimensional scaling as an alternative indirect ordination method used in ecology.
I intent to use the metaMDS function from the vegan package in R, which gives me reasonable results with the standard settings. But there are some questions that I couldn’t find clear answers to:
The help says that metaMDS is designed to be used with community data. Is in this context only biotic data meant? What would other data in this context mean? And what would my soil data fall into?
In case my soil data is not community data a range of settings has to be changed. This includes the dissimilarity index. As I am dealing with abiotic data, I generally would assume a linear gradient (as opposed to unimodal), should I in this case chose euclidean distance as dissimilarity index?
rankindex(env, soil, c("euc","man","bray", "jac", "kul"))
gives me the lowest score for euclidean (man and bray the highest).
EDIT I found several publications that point towards using the Gower metric dissimilarity index for soil chemical data. See the following book page 320 for more info: Legendre, Pierre, and Loic FJ Legendre. Numerical ecology. Vol. 20. Elsevier, 2012.. END EDIT
Do I have to change the name from nMDS to MDS if I use euclidean distance, as euclidean would be a metric method?).
Should the data be transformed before the calculations? Would autotransform = TRUE be sufficient?