An Investigation of the Relationship Between the Gaussian Process Model Parameters and Surface Area in Three Dimensions

In this paper, we study the relationship between the fitted parameters in a Gaussian process (kriging) model and the complexity of the resulting response surface. This study is done for models with one response and two input variables. An analytical calculation of surface roughness is used as a measure of the complexity of the response surface fit by the Gaussian process model. Our findings indicate that the size of the fitted model parameters as measured across different fits and data sets do not give indication as to the complexity of the surface. We do, however, show that the magnitude of each of the parameters in a single fitted model gives indication about the amount of variability in the direction of that fitted parameter.