A Statistical Approach for Estimating Fish Diet Compositions from Multiple Data Sources: Gulf of California Case Study

Trophic ecosystem models are one promising tool for providing ecosystem-based management advice. Diet and interaction rate parameters are critical in defining the behavior of these models, and will greatly influence any predictions made in response to management perturbations. However, most trophic ecosystem models must rely on a patchwork of data availability and must contend with knowledge gaps and poor quantification of uncertainty. Here we present a statistical method for combining diet information from field samples and literature to describe trophic relationships at the level of functional groups. In this example, original fieldwork in the northern Gulf of California, Mexico, provides gut content data for targeted and untargeted fish species. The field data are pooled with diet composition information from FishBase, an online data repository. Diet information is averaged across stomachs to represent an average predator, and then the data are bootstrapped to generate likelihood profiles. These are fit to a Dirichlet function, and from the resulting marginal distributions, maximum-likelihood estimates are generated with confidence intervals representing the likely contribution to diet for each predator–prey combination. We characterize trophic linkages into two broad feeding guilds, pelagic and demersal feeders, and explore differentiation within those guilds. We present an abbreviated food web for the northern Gulf of California based on the results of this study. This food web will form the basis of a trophic dynamic model. Compared to the common method of averaging diet compositions across predators, this statistical approach is less influenced by the presence of long tails in the distributions, which correspond to rare feeding events, and is therefore better suited to small data sets