2.4 Data statistics
According to Scott & Helfman (2001) and Liu et al. (2019), we identified native and native-invasive fishes for each fish species collected in this study. Specially, the native species with high degrees of endemism are typically adapted to cool, clear, lotic conditions in mountain streams, whereas the native-invasive species occurring mainly in large streams would be potential invaders of degraded headwater streams, which are highly adapted to warm, nutrient- and sediment-rich conditions (Scott & Helfman, 2001; Dala-Corte et al., 2019; Liu et al., 2019). The latest research has suggested that habitat degradation following human-induced disturbances may facilitate the native invasion of downstream fish fauna to headwater streams (Dala-Corte et al., 2019). From the fish data, the frequency occurrence (FO ) and relative abundance (RA ) were calculated for each species as FO i = 100(S i/S ) % andRA i = 100(N i/N ) %, where S i and S are the numbers of the samplings of which species i was collected and of the total samples, N i and N are the individual numbers of species i and of total fish species, respectively. Meanwhile, based on the number of species and number of samples (the reference sites and treatment sites), the species accumulation curves were constructed in R software (R Core Team, 2018) utilizing the “vegan ” package (Appendix S2).
Taxonomic similarity of fish assemblages was quantified by using the Bray-Curtis similarity index in this study. Specially, site ×species matrices were created separately for both the impoundments (Im ) and free-flowing segments (Fr ) based the abundance data (Figure 2a). Then, the taxonomic similarity was separately calculated separately for each habitat (impoundments:Im.TS ; free-flowing segments: Fr.TS ). Changes in pairwise taxonomic similarity (taxonomic ΔCS) were calculated between the impoundments and free-flowing segments, as follows: taxonomic ΔCS =Im.TSFr.TS (Figure 2a). For functional similarity, we first obtained the community-weighted mean of trait values per site (CWM) for both the impoundments and free-flowing segments using abundance data. The site × trait matrices (CWM) were created by multiplying the site × speciesmatrix and species × trait matrix for each habitat type (Figure 2b). The functional similarity was separately calculated for the impoundment assemblages (Im.FS ) and assemblages sampled in the free-flowing segments (Fr.FS ), according to the Euclidean distance (Dala-Corte et al., 2019; Daga et al., 2020). Following the same approach used for the taxonomic analysis, changes in pairwise functional similarity (functional ΔCS) were calculated as follows: functional ΔCS = Im.FSFr.FS (Figure 2b). Positive values for ΔCS indicated taxonomic or functional homogenization, whereas negative values indicated differentiation (Baiser & Lokweed, 2011; Pool & Olden, 2012; Daga et al., 2020). All the calculations were performed in R software utilizing the “FD ” package (Laliberté et al., 2014) and the “vegan ” package (Oksanen et al., 2013).
Using the log10 (x+1) transformed fish-abundance data, the one-way analysis of similarity (ANOSIM) was used to identify the discrete spatial variation in fish assemblages between the two habitat types. And the contribution of each species to differences among assemblage groups was assessed using the similarity percentages (SIMPER). These analyses were conducted in Primer 7.0 software. Paired samples t-tests were used to test the between-habitat (i.e., impoundments and free-flowing segments) differences in the mean taxonomic and functional similarity of fish assemblages.
Meanwhile, both the principal component analysis (PCA) and permutational multivariate analysis of variance (PERMANOVA) to test the differences in local habitat between the impoundments and free-flowing segments based on the standardized environmental variables. We further used the permutational analysis of multivariate dispersions (PERMDISP; Anderson, 2006) to test the differences in the environmental heterogeneity between the two habitat types. Using analysis of variance (ANOVA) F-statistics, PERMDISP compares among-group differences in the distance from observations to their group centroid, and the significance was tested via permutations of least-squares residuals (Heino et al., 2013). The Mantel permutation test was used to test the relationships between changes in taxonomic and functional similarity (taxonomic ΔCS and functional ΔCS) for abundance-based data. The significance of each Pearson correlation was calculated using the Mantel test with 9,999 permutations (Nekola & White, 1999). Differences in the observed percentages of each quadrant (i.e., homogenization or differentiation) were tested by Chi-square test with null hypothesis that the proportions were the same. The above analyses were performed in R software (R Core Team, 2018) utilizing the “vegan ” package (Oksanen et al., 2013).