Diagnostics for Gaussian Process Emulators

Summary: [This abstract is based on the authors' abstract.] Mathematical models known as simulators are widely used to represent complex real-world phenomena, but are so complex that they take considerable computer time and other resources to run. A methodology has been developed to build a statistical representation of the simulator. This emulator uses Gaussian processes. Diagnostics are presented to validate and assess the adequacy of the Gaussian emulator as a surrogate for the simulator. These diagnostics are applied to two different data sets.