Geostatistical Applications of Spartan Spatial Random Fields

Abstract

Spartan Spatial Random Fields (SSRFs) were recently proposed (Hristopulos 2003) as a new method for modelling spatial dependence. This paper focuses on (i) the inference of Gaussian SSRF model parameters from spatial data using kernel methods and (ii) the identification of geometric anisotropy by means of the covariance tensor identity (CTI) method (Hristopulos 2002). The methods presented are illustrated with the help of synthetic data and a real set of elevation data. Kriging predictions obtained with the Spartan covariance estimator are compared to those obtained with standard estimators. Based on these results, the Spartan estimator provides a useful alterative to parametric covariance estimators, which it may outperform in certain cases.