Kriging Temperature in Scotland using the External Drift Method

Abstract

Land use research includes the formulation and testing of models for land evaluation, land suitability and crop yield at the regional or district scale. Such models provide the bio-physical inputs necessary to make economic forecasts about the returns from alternative land uses. Modelling the seasonal and spatial variation of climatic variables is necessary for land evaluation and yield prediction. Temperature data are sparse in Scotland, with 149 climate stations covering a domain extending 600 km from north to south and 300 km from west to east. The temperatures are correlated with elevation, which is known everywhere, so it was possible to make use of this relationship to krige temperatures by using elevation as external drift. The underlying hypothesis is that the regression of temperature on elevation is linear. The spatial structures of monthly means of temperature were analysed by computing variograms in several directions. The temperatures are nonstationary in different directions in winter and summer but there appear to be stationary periods in spring and autumn. The underlying variogram of January temperature was used in universal kriging with elevation as external drift. The parameters of the linear regression with elevation may not be constant over the whole domain and using a moving neighbourhood allows for this. Kriging temperatures with elevation as external drift gives estimates of temperature which are closer to reality than those obtained kriging temperatures using only standard drift functions.