Development, Emulation and Analysis of Complex Computer Simulations with Real World Application Seminar

This talk is broken into two related parts, an outline of the applications and development of statistical methods performed within the HASP group of Dstl, focussing on outstanding problems of interest, will be followed by a specific research development; the increase of the flexibility of Design and Analysis of Computer Experiment methodology via the use of a thin plate spline basis function. More specifically, it is often desirable to build a statistical emulator of a complex computer simulator in order to perform analysis which would otherwise be computationally infeasible. A methodology to model multivariate output from a computer simulator taking into account spatial structure in the responses is described. The utility of this approach is demonstrated by applying it to a Chemical or Biological (CB) hazard prediction model which has been outlined in part one of the talk. The hazard area resulting from such a CB release is highly structured in space and we therefore propose the use of a thin-plate spline to capture the spatial structure and fit a Gaussian process emulator to the coefficients of the resultant basis functions. Three different techniques for emulating multivariate output are compared: a fully Bayesian approach with a principal component basis, a fully Bayesian approach with a thin-plate spline basis, assuming that the basis coefficients are independent and a \"plug-in\" Bayesian approach with a thin-plate spline basis and a separable covariance structure.