Comparing multivariate kriging methods for spatial estimation of air temperature

Abstract

In this paper three geostatistical algorithms are developed to incorporate exhaustive auxiliar information for the estimation of a primary attribute : kriging with varying local means and kriging with external drift, that use secondary data in order to characterize the primary spatial trend; and collocated cokriging, that explicitly account for the spatial cross-correlation between both attributes. The different methods are tested by means of air temperature data sets from the automatic network of the Meteorological Basque Service, incorporating elevation as explicative variable. Cross validation is used to compare the estimation performances. The main conclusion is that the application of multivariate geostatistical methods can improve the effectiveness of meteorological maps, even when relationship between temperature and elevation weakens.