Bayesian estimation of predator diet composition from fatty acids and stable isotopes
Quantitative analysis of stable isotopes (SI) and, more recently, fatty acid (FA) profiles are useful and complementary tools for estimating the relative contribution of different prey items in the diet of a predator. The combined use of these two approaches, however, has thus far been qualitative. We propose a mixing model for FA profiles that follows the Bayesian machinery employed in state-of-the-art mixing models for SI. This framework provides both point estimates and probability distributions for individual and population level diet proportions. Where fat content and conversion coefficients are available, they can be used to improve diet estimates. This model can be explicitly integrated with analogous models for SI to increase resolution and clarify predator-prey relationships.
We applied our model, fastinR, to simulated data to demonstrate feasibility and model performance. Testing on the simulated data confirmed that this approach provides more accurate and precise estimates of diet proportions than does analysis of SI or FA alone. However, simulation testing also highlighted the importance of prior information (e.g., from experimental feeding studies) about the conversion coefficients. We also used fastinR to re-analyze an experimental dataset to illustrate modeling strategies and applications to real SI and FA data. Posterior distributions of diet proportions using both SI and FA were substantially narrower than those obtained from using either SI or FA alone. Our methods are provided as an open source software package for the statistical computing environment R.