Statistical analyses
Generalized linear mixed-effects models (GLMM’s; Bates et al. 2015) were used to investigate how the (a) species richness, (b) abundance, (c) energy flux, and (d) relative energy flux (%) of whole-community and individual trophic compartments (top-carnivores, mesocarnivores, omnivores, and detritivores) changed over time (17 years). We nested season within stream sites (Nuevo Berlín, Fray Bentos, and Las cañas) as our random structure. We modeled species richness assuming Poisson-distributed errors as is common in count data. Biomass and energy flux were modeled assuming a negative.binomial distributed error to account for data overdispersion.
Linear mixed-effects models were employed to determine the relationship between energy flux and species richness, including whole-community and individual trophic compartments (top-carnivore, mesocarnivore, omnivore, and detritivore). We modelled the relationship between energy flux and species richness on a log-log scale because this specification has empirical supports in fish communities (Benkwitt et al. 2020). We nested season within stream sites (Nuevo Berlín, Fray Bentos, and Las cañas) as our random structure.
Structural equation models (SEM) were employed to address the direct and indirect pathways by which human pressure, precipitation, N:P ratio, and environmental variables affect the species richness and energy flux. We test in SEM the influence of time on the causal effects of drivers on diversity and energy flux (Fig. S5). We tested multicollinearity between drivers by calculating the variance inflation factor (VIF). VIF > 3 indicates possible collinearity, which was not observed in our model. We constructed SEMs for trophic guilds separately, hence, four SEMs were fitted: (i) top-carnivores, (ii) mesocarnivores, (iii) omnivores, and (iv) detritivores. The SEMs were fitted using a linear mixed-effect model in the piecewiseSEM package (Lefcheck 2016). We present the standardized coefficient for each path and estimated. We estimated the indirect effects of each driver on the energy flux mediated by species richness. Specifically, the indirect effect was estimated by multiplying the coefficient of each driver on richness by the coefficient of richness on energy flux. The significance of all paths was obtained using maximum likelihood and SEM fit was examined using Shipley’s test of d-separation through Fisher’s C statistic (p > 0.05 indicates an adequate model). All analyses were conducted in R 3.4.4 (RStudio Team 2020).