Statistical analyses
Canonical Correspondence Analyses (CCA) were used to explore differences in drift and stranding propensity, respectively, among the three HP reaches and to assess their correlation with a pre-selection of explanatory environmental variables (based on the state of knowledge, see introduction). Drift: Qpeak, Qpeak - Qbase, URmean, URmax, v0, vnet, v100, τb, Fr, T, NTU . Stranding:Qpeak, Qpeak/Qbase, DRmean, DRmax, v0, vnet, v100, DAex, T, NTU . Drift and stranding propensity, respectively, were used as response variable. Environmental variables were log-transformed (x + 1) and then sequentially checked for variance inflation factor (vif), while variables exceeding a vif of five were removed prior to the CCA modeling. The final selection of vif validated variables was based on a stepwise model selection (permutational forward and backward) based on the Akaike information criterion (AIC). Environmental variables that were selected in both approaches were used for the final CCA model construction. The CCA model was then tested for total significance, significant axes and significant terms via PERMANOVAs. An additional factor fitting was performed to assess statistical differences in drift and stranding propensity, respectively, between HP reaches (Sitter vs Hasliaare vsLinth), sampling day (day 1 vs day 2; corresponding to higher peak flow velocities at day 2, see Table 1) and HP scenario (SC 1vs SC2 vs SC3).
Non-metric multidimensional scaling (NMDS) was used to explore differences in the benthic community compositions among the different HP and RF reaches. Benthic densities were used to create a resemblance matrix based on the Bray-Curtis similarity index. Permutational multivariate analysis of variance (PERMANOVA) with a similarity percentage (SIMPER) analysis were performed to assess for significant differences in the benthic community composition among reaches and to identify the taxa that most contributed to dissimilarity.
In addition to CCA and NMDS, non-parametric Mann-Whitney or Kruskal-Wallis tests followed by pairwise post hoc tests, and linear regression models were applied for specific analyses. CCA and NMDS analyses were performed using the software R version 4.1.1 (R Core Team, 2021) with the package ‘vegan’ (Oksanen et al., 2019). Non-parametric tests and linear regressions were performed using the software SPSS Statistics version 27.0 and SigmaPlot version 12.5.