A NETWORK-BASED APPROACH FOR MODEL INTEGRATION

Models are simplified summaries of our understanding of the complex natural world. Simplifications are generally achieved through aggregations at various spatiotemporal scales and processes. The need for model integration arises when individual models built for limited spatiotemporal scales are to be combined to model at a less aggregated level. This talk will focus on the integration of various empirical models developed using localized data to form a regional model with an emphasis on modeling for supporting environmental management decision-making. The framework for integrating empirical models is based on conditional probability principles of a statistical graphic model, where components represented by various “small” empirical models are connected through a causal network. A network based model can be easily implemented using Markov chain Monte Carlo simulation. This talk will illustrate the process of model integration using a simple data example, which will lead to a general discussion.