Bayesian mixing models have become common in ecology, notably for tracking tracer molecules (stable isotopes and fatty acids) through food webs. A variety of options are available for generating mixing models, including IsoSource, SIAR, and MixSIR. We asked if these models can lead to misinterpretation of mixtures, or ecologically erroneous output, as a result of model parameterization. Mixing models are dependent on the trophic enrichment factor (TEF), also referred to as trophic discrimination factor (TDF) or diet-tissue discrimination factor (DTDF), assumed for each source. Typically, researchers use standard fractionation values from the literature for nitrogen and carbon isotopes in their models, regardless of diet item. However, there is evidence supporting variation in TEFs depending on the type diet item, consumer tissue type, and variation within a consumer taxon depending on the ecology and physiology of the population being studied. We used an isotopic dataset from Eastern Mosquitofish (Gambusia holbrooki) along with a simulated randomized dataset of source values to evaluate the impact of TEFs reported in the literature tied on the diet item (plant/algal, invertebrate, or a high protein food) and consumer tissue type (muscle or whole body) on model output and model fit.
Results/Conclusions
Our simulations revealed mixing model output is sensitive to the TEF, potentially altering the contribution of a source item to the consumer’s isotopic composition by up to 19.5%. This was the difference in percent contribution to the consumer diet for the same food source when an invertebrate-specific TEF was used (11.2% contribution to the diet) versus when a high-protein TEF was assumed (30.7% contribution to the diet). Model output was less sensitive to analysis of whole body versus muscle TEFs (0.7% to 3.7% change in contribution of a source to the consumer diet). TEFs with lower stand deviations led to increased model fit based, indicating sample size should be optimized through a power analysis to improve confidence in results. These results support using default TEFs specific to each diet item. Further studies are needed to experimentally obtain TEFs for diet items and to examine assimilation and fractionation of isotopes under various conditions. Future work with this simulated data set will use information theory to assess how TEFs impact information gain as well as how information theory can be used to assess the validity of a TEF in a model.