Bayesian inference for bioenergetic models

Dynamic energy budget (DEB) theory provides a sophisticated, mechanistic framework for understanding the full life cycles of individuals within a complex environment. By relating environmental conditions, notably food availability, to individual life histories, DEB theory makes it possible, in principle, to make predictions for individuals and populations that extend beyond the conditions used to develop and parameterize the model. However, for this approach to reach its full potential as a predictive theory, we need methods of similar sophistication to link unobservable modeled quantities to data and to infer model parameters. Here, we develop such a method in a Bayesian framework. We offer a rigorous methodology for modeling the link between underlying unobservable states and observable quantities and for parameter inference. The methodology is introduced and its effectiveness as applied to data simulated under various dynamic food regimes is demonstrated. We also examine how parameter estimates can be affected by misspecification of the food model, specifically by assuming a constant food level when the true underlying dynamics are variable. This is critical as many published applications of DEB theory assume constant food when estimating parameter values. The effectiveness of the approach for data on the growth and reproduction of the water flea Daphnia is also discussed.