IAMG 2001--Cancún Technical Program--Session K

Home Page Technical Program Session K Intro Session K Papers

Using Linear and Non-linear Kriging Interpolators to Produce Probability Maps by Konstantin Krivoruchko, Environmental Systems Research Institute, Redlands, California, USA Probability maps are used to define areas with high and low certainty of exceeding selected threshold value. The most popular methods for creating probability maps are variants of indicator kriging and conditional sequential simulation techniques, which have obvious disadvantages. For example, the use of such methods is questionable in the presence of trend in the data and in the case of measurement errors, common in the analysis of real data. In this presentation we are considering possible alternatives to mapping the risk of exceeding specified critical value using simple, ordinary, universal, and disjunctive kriging with data detrending and data transformation options (i.e. so-called kriging on residuals, lognormal and trans-Gaussian kriging). Creation of probability and quantile maps requires assumption regarding the multivariate normality for ordinary, simple, and universal kriging. Bivariate normality is assumed to obtain probability maps using disjunctive kriging. The diagnostic tools for checking univariate and bivariate normality provide informed decision about the possibility to use linear kriging methods for probability mapping. Case study using environmental data and comparison of the prediction results for different kriging methods will be presented and discussed. Paper in PDF format Krivoruchko, Acrobat PDF, 1.1 megs.