Describing the Feeding Ecology of Piscivorous Fish Using a Hierarchical Community Approach

We developed a Bayesian hierarchical community model to account for imperfect detection of prey types in predator stomachs by separately estimating occurrence and detection, which are often confounded.  Occurrence describes if prey type j is present in the population being sampled while detection is the probability of observing prey type j in the predator stomach sample, given it occurs in the sampling area.  Occurrence has a Bernoulli prior with the success parameter p defined as occurrence probability.  Occurrence probability describes the proportion of the samples where prey type j occurs.  Detection is modeled using an uninformative Beta hyperparameter.  Data augmentation is used to estimate occurrence and detection for prey types not detected during sampling, allowing for an estimate of the total number of missed prey types.  The model is flexible and can be used to estimate the total number of prey types in a predator’s diet, determine the required sample size to fully describe the diet, and describe how the diet varies in relation to abiotic and biotic factors.  Given the mechanistic understanding of diet, the model has potential for widespread use in future food habits studies.