Community trends: RLQ analysis
Trait and community responses to environmental gradients were ordinated using an RLQ framework, a multivariate technique that summarizes joint structure among matrices (Dray and Legendre 2008; Dray et al. 2014). In our case, these three matrix tables L (species distribution across river reaches surveyed as 15 species abundances*68 sites), R (environmental characteristics of samples: 68 sites*12 environmental variables) and Q (species traits: 15 species*8 traits) were analysed separately using different ordination methods in “ade4” package in R (Dray and Dufor 2007). The L-species table was analysed using Correspondence Analysis (CA), while the R-environmental variables table and Q-trait table were analysed by a Hill-Smith PCA combining quantitative and qualitative variables using CA species scores as a column weight to couple Q and L (Brown et al. 2014). In trait analysis, the RLQ approach crosses traits and environmental variables weighted by species abundances with significant effects tested using a two-step permutation procedure (25000 permutations). Model 2 permutes the rows of the L matrix to test the null hypothesis that no relationship exists between species abundance data with fixed traits and their environment; model 4 permutes the columns of dataset L to test the null hypothesis that species composition is not influenced by species traits given fixed environmental characteristics (Dray et al. 2014).