THE USE OF SPATIAL STATISTICS AND NUISANCE VARIABLES IN ESTIMATING PARAMETERS FOR VALIDATING MODELS AND PREDICTIONS

摘要

Estimation of parameters for nonlinear systems is often computationally expensive and slow to converge. The procedure is aggravated when parameters, other than those sought, are not known with precision. By using Bayesian estimation, it is possible to eliminate such nuisance variables, but often at considerable cost. Using response surface approaches is effective only a) for linear systems, b) when the initial guess is close to the final values. By using concepts from spatial statistics and incorporating nuisance variables, the convergence is improved and only a reduced number of computations need be performed. Spatial statistics also provides an efficient means for predictions.