Author:

Keywords:

statistical emulation, metamodeling, surrogate modeling, computational simulation, computational experiments

Abstract:

Statistical emulation is a technique for studying the behavior of computational simulation models. With this approach, a statistical function is fitted to the observed relations between model inputs and outputs based on systematic experimentation with the simulation model. The resulting function provides information about the general behavior of the simulation model and can be used, for example, for model simplification, optimization, and calibration. In this article, we discuss the general principles of statistical emulation, introduce readers to regression metamodels and Gaussian process emulators as two examples of commonly used statistical functions, and point readers to experimental designs that can be used for fitting these types of functions.