Non-parametric multivariate regression tests (dbRDA) for relationships between habitat transformation gradients and physico-chemical conditions in wetlands.

Natural - indigenous vegetation; Invaded - alien invasive vegetation; Agriculture - agricultural land; Urban - urban area. The areal cover of these variables is represented within 100 and 500 m radii of each wetland edge. To maximise parsimony, covariable subsets were pre-selected for each model using step-wise regression of each response matrix on the full list of possible covariables (see Table 1). % Var - the percentage of variation in each Euclidean distance matrix (normalized physico-chemical variables) that is explained by each respective predictor variable or covariable set in each model; Time – number of days since the first sampling event; SF – Sand fynbos; SR – Shale renosterveld; FF – Ferricrete fynbos; Res. df – residual degrees of freedom for each model. Significant P values at α<0.05 (*), α<0.01 (**) and after sequential Bonferroni correction (***) are indicated.