Contributions of body size, habitat and taxonomy to predictions of temperate Australian fish diets

ABSTRACT: Using k-nearest neighbour procedures to predict prey type and linear models to predict mean prey size, we developed a 2-step dietary model based on the stomach contents of fish of known species, size and location from Western Port, Victoria (Australia). The model, nicknamed ‘Consume’, was used to assess the relative extent to which fish diet varied with body size, species identity, season, and location. Both prey type (mean overlap between predicted and actual prey types = 77%) and mean prey size (r2 between predicted and observed mean prey size = 93%) were predicted with reasonable accuracy when species identity and length of consumer fish were known. The most important predictor for prey type was the size of the individual consumer, while the most important predictor for mean prey size was the consumer’s taxonomic identity. Predictors were individually removed from both k-nearest neighbour and linear models to assess their relative contributions to the model. Little loss of accuracy (1%) was evident when family rather than species identity was used for both prey type and mean prey size. Environmental information associated with the time and location of fish sampling (habitat, site and season) contributed only marginally to predictions of prey type. Use of the Consume model will allow for an improved understanding of community-level trophic pathways through the integration of prey type and size predictions for consumer fishes.