Local and Spatial Uncertainty Models Based on Ordinary Kriging
Linear kriging aims at mapping a regionalized variable on the basis of a sample on this variable. The mean value is often regarded as an unknown parameter (ordinary kriging) that can vary from one neighborhood to another. So far, the use of ordinary kriging has been extended to nonlinear techniques such as indicator, disjunctive and multi-Gaussian kriging, in order to estimate conditional distributions and predict transfer functions. However, these techniques do not yield accurate uncertainty measures in the presence of an unknown mean value. This article proposes a flavor of the multi-Gaussian model, in which the original data are transformed into normal scores whose mean value is regarded as a random variable constant in space instead of a deterministic parameter. It is shown that, if the prior variance of the random mean is infinitely large, men the conditional distributions of the normal variable are obtained by replacing simple kriging by ordinary kriging in the multi-Gaussian formalism. The results are used to calculate the conditional distributions at a single location (point-wise uncertainty models) or at multiple locations jointly, and to perform conditional simulation. The conditionings by simple and ordinary kriging are compared through a case study.
Xavier Emery
Department of Mining Engineering, University of Chile, Avenida Tupper 2069, Santiago 837 0451, Chile
国际会议
The 12th Conference of the International Association for Mathematical Geology(第12届国际数学地质大会)
北京
英文
565-568
2007-08-26（万方平台首次上网日期，不代表论文的发表时间）