|
Abstract:
|
Emulators, by providing computationally efficient interpolation between outputs of deterministic simulation models, can considerably extend the range of statistical techniques that can be applied to computationally demanding models. So far, the dominant technique for developing such emulators were based on Gaussian stochastic process priors. In the context of dynamic models, this approach has two essential disadvantages: (i) these emulators do not consider our knowledge of the structure of the model, and (ii) they can run into numerical difficulties if the output covers the time dimension densely. To address these problems we propose an emulator that is based on a simplified linear state-space model with Gaussian stochastic processes of the innovation terms as functions of model parameters and/or inputs. Conditioning this prior to the design data set is done by Kalman smoothing.
|
