Abstract:
|
Computational models are frequently used to explore physical systems. Often, there are several computer models (or simulators) available and also a limited number of observations from the physical system. The aim is then to combine the simulators' output and field data to build better a predictive model of the system of interest and also estimate parameters that govern the system. Here, new methodology is proposed for combining multi-model ensembles of outputs and field observations to make predictions for the system with uncertainty. For the proposed approach, we do not choose a "best" model, but instead use a spatially varying convex combination of simulators. The methodology is motivated by an application in glaciology.
|