Boosted regression tree models predict the diets of juvenile bull sharks in a subtropical estuary

Abstract

Understanding diet flexibility is important for resource management as climate change alters ecological communities. However, food web complexity often limits our ability to predict how changes in prey communities may alter predator diets. Stomach content and stable isotope analyses are traditionally used to evaluate trophic interactions, but costs and logistical constraints can limit their efficacy. Using boosted regression tree (BRT) models, we predicted how juvenile bull shark Carcharhinus leucas diets respond to shifts in potential prey communities using patterns of shark and prey distributions, and size-based differences in shark gape widths. BRT models were based on bull shark diets from published literature and long-term monitoring of sharks and prey in a coastal estuary in the western Gulf of Mexico. In situ diet data were used to test model accuracy, which revealed that BRT models effectively predicted the most abundant prey families in the diets of bull sharks (Sciaenidae: ~37%; Ariidae: ~34%), with Pearson’s correlation rates as high as 0.778 for predictions and in situ diet data. Inaccuracies were evident for rarer prey families (e.g. Mugilidae), which was attributed to monitoring limitations, elucidating how BRT models can be improved before future application. High model accuracy suggests BRTs may serve as an appropriate complement to stomach content and stable isotope analyses when monitoring data of predators and potential prey are available. Such results are promising for reducing stressful or harmful sampling and broadening the applications of current monitoring programs used to assess changes in species densities and distributions, particularly for resource-limited management agencies.