A powerful aspect of trait-based approaches to ecology is the ability to predict future species composition. Typically, these predictions require some method to estimate community-aggregated traits (CA-traits) of the species in the new environment. Under community assembly theory, CA-traits can be estimated from regressions against habitat variables that act as strong filters to species presence/absence. However, CA-traits could also exhibit spatial structure; for example, similarity between communities often decreases with distance. Here, we estimate CA-traits using habitat regressions and spatial analysis, and then compare the ability of these methods to predict CA-traits. Within a 700-m reach of a second-order stream, we sampled four habitat variables (depth, flow, sediment size, and riparian canopy cover) and the abundance of stream fish species at two time intervals (July and October). For July, five CA-traits (body length, trophic position, caudal fin aspect ratio, reproductive guild and substrate preference) were calculated at each fish sampling point. CA-traits were then regressed against the habitat variables and the resulting regressions were used to predict the CA-traits of fish sampled in October. To detect spatial structure, we constructed correlograms for each CA-trait. We then interpolated CA-traits throughout the stream and used these values to predict October CA-traits.
Significant regressions were found for each CA-trait and the strongest relationships were between body length and water depth (slope ± 1 SE; 0.192 ± 0.039, r2 = 0.437, P < 0.001) and substrate preference and sediment size (0.303 ± 0.064, r2 = 0.408, P < 0.001). Only one CA-trait (trophic position) exhibited any evidence of spatial structure (Moran’s I = 0.142, P = 0.003). For October, body length (r2 = 0.397, P < 0.001), aspect ratio (r2 = 0.271, P = 0.001), and substrate preference (r2 = 0.172, P = 0.013) were best predicted using the habitat regressions, while trophic position (r2 = 0.033, P = 0.296) and reproductive guild (r2 = 0.162, P = 0.017) were best predicted using interpolation. Habitat seems to be more important than distance in structuring stream fish assemblages at a local scale. Most CA-traits did not exhibit strong spatial dependence, and predictions of CA-traits for the October fish assemblage were much stronger using the habitat regressions. The lack of spatial structure in stream fish assemblages may be due to the high degree of habitat heterogeneity observed in these ecosystems and the adaptations of fishes to these habitats.