Towards Identifiable Dynamic Stochastic Models in Ecology and Beyond – TIDySModels

Identifying the parameters of dynamic models from observed state variables is a crucial but challenging endeavour, especially for interacting biological systems, such as interacting populations of different species. Nonlinear functions, stochasticity induced by both biological processes and measurement error, and a moderate information content in field-based population sizes combine to render inverse estimation hazardous. In fact, for many seemingly standard models, we do not know whether parameters are truly identifiable (i.e., if a unique parameter set or single-peaked distribution can be defined), both in theory (infinite dataset) and practice (with limited data). TIDySModels will diagnose when dynamic models are identifiable or not, and will improve parameter identifiability by smart combinations of multiple data streams. Two ecological examples of practical importance are used: quantifying regulation by predators, as well as explaining how biodiversity maintains.