Kriging with large data sets using sparse matrix techniques

This work compares the relative efficiency of variogram-based and covariance-based kiiging, using both real and simulated data.

Abstract

A major impediment to kriging with large data sets is the need to solve matrix equations with the large matrices that result from using variogram-based kriging equations. This is expensive both in computing time and memory. When the range of the variogram is small, use of covariance-based kriging equations and sparse matrix techniques can allow the kriging equations to be solved very efficiently. By fitting the variogram model and then using this to derive the covariance matrix, we keep the better estimation properties of the variogram, and can exploit the sparseness of the covariance matrix. We compare the relative efficiency of variogram-based and covariance-based kiiging, using both real and simulated data. We also comment on the use of sparsity in kriging