An empirical test of fatty acid based diet estimation models

Due to the perceived power of including many variables, models tracing predator prey relationships with fatty acid profiles have recently garnered much interest. Currently, Bayesian mixing models (FASTAR and mixSIAR) and a numerical optimization model (QFASA) have been established through various simulations and case studies. To date, validation has yet to be attempted through controlled feedings of both monotypic prey species as well as mixed composition diets to fishes. As such, we evaluate which model provided the most accurate diet estimates by feeding Lake Trout known mixtures of prey for up to 12 weeks. Visualizations of Lake Trout fatty acid profiles using nMDS or linear discriminant analysis indicated distinct fatty acid profiles for each of the seven diet-treatment groups, with profiles from mixed diet treatments within a resource polygon outlined by those from monotypic diet treatments. Despite optimistic ordinations, all three diet estimation methods estimated diet compositions with errors of 7.3–35.0% per prey item per Lake Trout. Errors did not seem to improve with greater feeding time, increases in length, mass, or lipids of Lake Trout, nor with a reduced suite of fatty acids. We cannot rule out the effects of malnutrition or differential nutrient offerings on consumers' fatty acid profiles but argue nor could one with wild organisms. Fatty acid profiles were extracted from whole body samples, and the inclusion of organs and neural tissues may also have influenced results. Further evaluation of such models is suggested prior to utilization for management decisions.