Benthic assemblages are more predictable than fish assemblages at an island scale

Abstract Decades of research have revealed relationships between the abundance of coral reef taxa and local conditions, especially at small scales. However, a rigorous test of covariation requires a robust dataset collected across wide environmental or experimental gradients. Here, we surveyed spatial variability in the densities of major coral reef functional groups at 122 sites along a 70 km expanse of the leeward, forereef habitat of Curaçao in the southern Caribbean. These data were used to test the degree to which spatial variability in community composition could be predicted based on assumed functional relationships and site-specific anthropogenic, physical, and ecological conditions. In general, models revealed less power to describe the spatial variability of fish biomass than cover of reef builders (R 2 of best-fit models: 0.25 [fish] and 0.64 [reef builders]). The variability in total benthic cover of reef builders was best described by physical (wave exposure and reef relief) and ecological (turf algal height and coral recruit density) predictors. No metric of anthropogenic pressure was related to spatial variation in reef builder cover. In contrast, total fish biomass showed a consistent (albeit weak) association with anthropogenic predictors (fishing and diving pressure). As is typical of most environmental gradients, the spatial patterns of both fish biomass density and reef builder cover were spatially autocorrelated. Residuals from the best-fit model for fish biomass retained a signature of spatial autocorrelation while the best-fit model for reef builder cover removed spatial autocorrelation, thus reinforcing our finding that environmental predictors were better able to describe the spatial variability of reef builders than that of fish biomass. As we seek to understand spatial variability of coral reef communities at the scale of most management units (i.e., at kilometer- to island-scales), distinct and scale-dependent perspectives will be needed when considering different functional groups.