An Intelligent Improvement on the Reliability of Ordinary Kriging Estimates by a GA

Abstract

One of the most common models used to define the OK neighborhood, a search ellipsoid, has some shortcoming. For instance, it can determine values of the resulting estimates but itself is strongly dependent on the user’s knowledge to some extent. This paper presents a smart search ellipsoid (SSE) model to improve the reliability of OK estimates. By constructing an evaluation criteria mainly consisted of the kriging variance and interpolation variance, a participant sample set within the common search ellipsoid (CSE) model can be refined intelligently by a genetic algorithm process. In theory, the output from SSE can guarantee the least estimation uncertainty measured by both kriging variance and interpolation variance. Cross-validation was employed to investigate the interpolation performance of the proposed method. It can result in that OK estimates with the improved model have more global and local accuracy than before. What is noteworthy is that the proposed SSE model does not change the traditional procedure in modeling with OK but only improve the reliability of the estimates intelligently. So this method is valuable, especially for the practitioners in geo-statistics who will mostly be confused by the CSE parameters.