Spatial modeling and interpolation of monthly temperature using kriging

Estimates of the climatic record at forest plot locations may be useful in studying how forests will respond to future climatic change. Kriging was applied to the spatial interpolation of monthly temperature records in the forested regions of Minnesota in the north central United States. Monthly empirical variograms, averaged over 90 yr, were modeled with Gaussian or linear models, and ordinary kriging was applied to interpolate the data. Anisotropies were found in the winter months, suggesting the presence of a large-scale regional trend. Structural analysis of mean monthly ternperature revealed: (1) a broad regional component to the variation, changing systematically by month, which was estimated by a Linear function of latitude and longitude and (2) a lake effect (due to Lake Superior) varying in strength and sign by month. A detrending approach was tested to remove these effects and a modified approach based on only the lake effect trend was also tested. The cross validation technique was used to test the 3 models. The lake effect trend model was judged best in accounting for the influence of Lake Superior on nearby land areas. The study demonstrated that moderate trends in the data do not seriously degrade the applicability of ordinary krigmg to the interpolation of monthly temperature. A temporal analysis revealed that, although there have been systematic changes in the spatial variabhty over the last century, using century-averaged variograms is not expected to decrease the accuracy and precision of the interpolations.