Spartan models of spatial dependence
Abstract
Modelling the variability of spatially distributed data often involves the classical geostatistical framework, which requires calculating two-point variogram functions that characterize the spatial dependence. This is a computationally intensive procedure, especially for large-size samples. In addition, calculation of the variogram from a single sample realization relies on a number of assumptions. We propose an alternative method of modelling spatial dependence, which is based on random fields that we call Spartan, because their probability density function is determined from a small number of parameters. We present some general properties of Spartan random fields, and we further investigate specific models. We also present a specific algorithm for inferring the field parameters from available samples. The algorithm is illustrated with the help of synthetic samples, both with regular (lattice) and irregular (random) spatial distribution. The advantage of the Spartan models is the numerical efficiency of the model inference process, which is considerably faster than the standard variogram calculation.
- Publication:
-
EGS - AGU - EUG Joint Assembly
- Pub Date:
- April 2003
- Bibcode:
- 2003EAEJA.....8299H