We propose an extension of a parameter-free loss function for the estimation of Dynamic Energy Budget parameters for a set of related species that is symmetric in the role of data and predictions. The extension allows that particular parameters might vary, but not by much, among species, while the degree of variation is controlled by weight coefficients. We discuss the choice of these coefficients and illustrate the application with an example of two species of catfish, with their mutual hybrids. In our simultaneous parameter estimation for this example, we could reduce the variation in a parameter, here the energy conductance, substantially with a minor effect on the goodness of fit. We selected this parameter among the ones that varied, because its value was poorly determined by the data. We discuss this example in some detail. The software for implementing this augmented loss function has been implemented in the available software DEBtool_M on Github (www.github.com/add-my-pet/DEBtool_M).
- Clarias & Heterobranchus
- Dynamic energy budget theory
- Parameter estimation in context
- Regularization of parameter estimation
- Symmetric loss functions