A Bayesian approach to identifying mixtures from otolith chemistry data

Studies investigating population structure and mixed-stock composition of fish populations frequently use otolith chemistry as a natural tool for discerning stocks. Current methods for estimating mixed-stock composition, however, assume complete accuracy in the training data, which is often not the case. Here we present a method for estimating mixed-stock composition using multivariate continuous data that accounts for uncertainty in the training data. Application of the method to previously reported data for natal homing in weakfish (Cynoscion regalis) and simulations based on these data revealed that for sample sizes greater than about 30, the present method provides results that are quite similar to those of previous methods. An advantage of this Bayesian approach over other methods, however, is the ease with which functionals of the model, such as migration distance and direction, can be calculated. It also provides simple means of visualizing spatial structure in the classification probabilities and migration patterns.