The verification of numerical models with multivariate randomized block permutation procedures |
| |
Authors: | D F Tucker P W Mielke Jr E R Reiter |
| |
Institution: | (1) Present address: Center for Cybernetic Communication Research, Colorado State University, 80523 Fort Collins, Colorado, USA;(2) Present address: Department of Statistics, Colorado State University, 80523 Fort Collins, Colorado, USA;(3) Present address: U.S. Army Atmospheric Sciences Laboratory, White Sands Missile Range, New Mexico, USA |
| |
Abstract: | Summary Multivariate randomized block permutation procedures (MRBP) can be used effectively to verify numerical models. Compared to other statistical methods, MRBP shows several distinct advantages. First of all, MRBP operates in the same Euclidean analysis space as its input data. The root mean square error (RMSE) is discussed, since it is a natural choice as a distance measure between two data sets and is closely related to the distance measure on which MRBP is based. The RMSE by itself provides no basis for inferential comparisons, whereas MRBP is well suited for such deductions. Since MRBP is computationally economical and requires only a few case studies for meaningful comparisons, it is also useful for model development.With 3 Figures |
| |
Keywords: | |
本文献已被 SpringerLink 等数据库收录! |
|