2.3 Data analysis
We used multivariate techniques to examine the influence of connectivity and seasons on amphibian, fish, benthos, and plankton assemblages, respectively, and also all biota together. Density data of all faunal groups were standardized using the decostand(XX, method=”max”) function in the vegan package (Oksanen et al., 2020) in R prior to analysis so that the density data of each taxon range between 0 and 1. We then analysed the assemblage data at 25 sites in three seasons (as 78 data) by nonmetric multidimensional scaling (NMDS) using Bray-Curtis dissimilarity index as pairwise β-diversity values. We applied two-way PERMANOVA procedures to the β‐diversity values to test for statistical differences in the biota by the connectivity gradient and by seasons using the adonis function in the vegan package (Oksanen et al., 2020). When there were significant interactions of connectivity and seasons, we applied PERMANOVA for the 25 data of each season to examine the effect of connectivity on the assemblages in each season.
We then used generalized dissimilarity modelling (GDM) to analyse the relative contribution of connectivity and local environment on the β-diversity in each season. GDM is a nonlinear extension of matrix regression for analysing and predicting patterns of assemblage dissimilarity in relation to environmental gradients. Dissimilarities between all possible pairs of assemblage were calculated and were modelled as a function of environmental dissimilarities assuming their nonlinear relationships (Ferrier et al., 2002, 2007). We applied GDM to 25 assemblage data for each season. Using the gdm package (Manion et al., 2018), we first converted each dissimilarity matrix and the environmental data to site-pair format, applied GDM with function gdm, and estimated importance of environmental data using function gdm.varImp. As connectivity variable, we used the four connectivity categories “No flow”, “Early”, “Late”, and “Flowing” based on the presence of inflow from the main stream over time as described above. As environmental variables, we included area and depth of each pond, average flow velocity, downstream connectivity of each pond to the river, grain size, water temperature, and dissolved oxygen. pH and conductivity were not included for the analysis because of their high association with DO (positive) and temperature (negative).
To partition variation in biotic assemblages, each group of explanatory variables was first screened with forward selection with a Monte-Carlo randomisation test and only variables significantly related to assemblage structure in forward selection were retained in the final models. We then carried out three model fittings for each assemblage data set along with a different set of environmental data matrix. Matrix-(I) was constrained by both connectivity and local environments (a +b +c ; where a describes pure effects of environmental variables, b shared environmental and connectivity effects, and c pure effects of connectivity); matrix-(II) was constrained by environmental variables only (a +b ); and matrix-(III) was constrained by connectivity only (b +c ). Variation in assemblage structure was subsequently partitioned into shared environment and connectivity [b =(a +b ) +(b +c ) −(a +b +c )], pure local environment [a =(a +b ) −(b )], pure connectivity [c =(b +c ) −(b )], and unexplained fractions [d =1 −(a +b +c )] (Borcard et al. 1992; Legendre and Legendre 1998). In addition, in order to illustrate autumn benthic macroinvertebrate and amphibian assemblage patterns in relation to local environments (Table S1) a canonical correspondence analysis was conducted.
Finally, to identify key taxa driving the spatial and temporal variations in aquatic assemblages, we applied a similarity percentage analysis (SIMPER) to examine the contribution of each taxon to the differences in the whole aquatic assemblage of each season. We also carried out SIMPER to examine the seasonal change in aquatic assemblages and tested the significance in the seasonal decline or increase of each taxon from spring to autumn.