UW Aquatic & Fishery Sciences Quantitative Seminar
UW Quantitative Ecology & Resource Management
Multi-objective optimization for ecological process model assessment
Individual and process-based models are increasingly utilized to simulate complex ecological systems, particularly in the early stages of investigation. Often the goals of such models are to reconstruct the system of interest through analysis of the emergent properties of the simulated system. These models are complex and produce multiple outputs that require a rigorous method to assess the process structure and individual interactions, and the mathematical formulations used to represent these. Deficiencies in model structure choices may be masked during traditional univariate model assessment (e.g. maximize R2) due to the severe loss of model performance information imposed by a univariate assessment function. Assessment of a process model against a vector of multiple objectives, each measuring different features of model performance, provides a method by which deficiencies of model structure can be discovered and the model modified accordingly. This is accomplished by approximating the model's Pareto frontier, the set of non-dominated parameter solutions. I present an example of this methodology for a model of the branching system of old-growth Douglas-fir (Pseudotsuga menziesii).