Evaluating the performance of the Bayesian mixing tool MixSIAR with fatty acid data for quantitative estimation of diet

We test the performance of the Bayesian mixing model, MixSIAR, to quantitatively predict diets of consumers based on their fatty acids (FAs). The known diets of six species, undergoing controlled-feeding experiments, were compared with dietary predictions modelled from their FAs. Test subjects included fish, birds and mammals, and represent consumers with disparate FA compositions. We show that MixSIAR with FA data accurately identifies a consumer's diet, the contribution of major prey items, when they change their diet (diet switching) and can detect an absent prey. Results were impacted if the consumer had a low-fat diet due to physiological constraints. Incorporating prior information on the potential prey species into the model improves model performance. Dietary predictions were reasonable even when using trophic modification values (calibration coefficients, CCs) derived from different prey. Models performed well when using CCs derived from consumers fed a varied diet or when using CC values averaged across diets. We demonstrate that MixSIAR with FAs is a powerful approach to correctly estimate diet, in particular if used to complement other methods.