Results/Conclusions
The location of waterfalls was the best predictor of most species distributions and assemblages, with all predatory fishes limited to areas below waterfalls and greater shrimp species richness and abundance above waterfalls. Local pool habitat was less important in predicting species assemblages, but pool length and grain size best predicted four species distributions. Given the strikingly disparate distributions of predatory fishes and the shrimp Atya lanipes, we conducted experiments in artificial and natural streams to test the role of predation in further influencing community structure. Using a y-maze with fish scent on one side, shrimp showed slight avoidance behavior in response to two fish species. We also set up flow-through tanks in four headwater streams and added fish to two of them. We marked shrimp with unique colors in pools above and below the fish scent input. We recaptured shrimp with the expectation that shrimp exposed to fish scent would move upstream to avoid predation. We did not observe a decline in shrimp abundance or a change in movement patterns in the streams where fish scent was added. However, shrimp abundances did decline dramatically when fish were put into wire-mesh cages placed directly in the stream. Diadromy is a major component of tropical island stream community structure, determining the species pool available to the island and the upstream extent of predatory fishes. Streams above waterfalls may provide refugia from predation for shrimps.