Coupled models: Expert Judgement, Emulators and Model Uncertainty

The project will focus on applications of the new technique of emulators to a number of closely related problems in risk assessment. An emulator is essentially a statistically built approximation for a model, which is much faster to run than the original model. It is common in risk assessment to chain complex computer models together to generate information about scenario's of interest to the decision makers. In this project the use of chained emulators will be studied with a number of research topics in mind. First is to explore how emulators can be used to support expert judgements about uncertainties, when experts have opinions about measurable quantities that might be observed at different stages of the model chain, so that we might obtain a better way of quantifying parameter uncertainties. Second is to explore how parameter uncertainty for a model depends on the modelling context in which it is used (in this case, the chain of models it forms a part of). The third and final question recognizes that there are often several different models that could have been chosen to model some kind of behaviour, and therefore seeks to explore the possibilities of using emulators from those different model alternatives to assess the structural sensitivity of a model chain to those modelling choices.