Accuracy of geostatistical prediction of yearly precipitation in Lower Saxony

Geostatistical methods allow the interpolation of spatially referenced data. They allow the prediction of values for arbitrary points in the area of interest. From these point predictions, maps are created which display areas of low or high predicted values. Apart from three dimensional and raster plots, contour line maps are frequently used. Contours are lines with constant process values. The article describes two accuracy concepts of geostatistical predictions. Local errors describe the accuracy of individual point predictions as mean squared prediction errors. They do not describe adequately the precision of contour lines extracted from an interpolated surface. A method is proposed to construct horizontal confidence intervals for arbitrary contours. This method is based on the theory of conditional simulations and is applicable whenever a suitable simulation algorithm is found. In a case study, two geostatistical models are compared regarding both accuracy concepts. Results show that higher local precision need not induce higher contour line accuracy. It is concluded that the local error concept is not sufficient to evaluate the precision of a contour line map.