Multivariate analysis

Relationships between the environmental and biological (S. damnosum larvae and blackfly predators) data were examined using multivariate analysis with the software program CANOCO for Windows version 4.5 [62]. Detrended Correspondence Analysis (DCA) was used to determine the appropriate response (linear or unimodal) for biological data. The performed DCA yielded a length of gradient greater than 2 standard deviations, implying that the biological data exhibit a unimodal type of response along environmental gradients. Therefore, we used Canonical Correspondence Analysis (CCA) for data analysis. Prior to data analysis, when two or more variables had a VIF greater than 5, one of these variables was removed from the analysis. AVIF of 5 and greater has been identified as an indicator of collinearity in multivariate analysis [63]. Biological and environmental data, except pH, were log-transformed [(log (x + 1)] to improve normality and homoscedasticity.

A stepwise forward selection was employed to identify the smallest set of statistically significant variables that contribute most to the explained variance in the response variables. The statistical significance of eigenvalues and taxa-environment correlations generated by the CCA were tested using a Monte Carlo test with 999 permutations [62].