Many environmental phenomena are regarded as realizations of random functions which possess both spatial and
temporal characteristics. In particular, Geostatistics with an extension of the existing spatial techniques into the
space-time domain offers some kinds of methods to model such processes. Although these methods for the analysis of
spatial-temporal data are becoming more important for many areas of application, they are less developed than those for
the analysis of purely spatial or purely temporal data. In this paper, two kinds of spatial-temporal stationary covariance
models are introduced. And the differences between spatial domain and time domain are examined. A product-sum
covariance model originally given by De Cesare is extended for spatial-temporal analysis on daily rainfall measurements
in the three provinces of Northeast China. Remarkably, this generalized non-separable model does not correspond to the
use of a metric one in space-time. The rainfall measurements used for this experiment are taken at 104 monitoring
stations from January 2000 to December 2005. In the experiment, the product-sum variogram model is employed for
developing ordinary kriging and its application to interpolation of the monthly rainfall data from January 2000 to
December 2004 has been used to predict the monthly rainfall of 2005. The true values and the predicted ones are
compared. The experimental results have shown that this product-sum covariance model is very effective for rainfall
analysis.
|