Use of cokriging to estimate surface air temperature from elevation

Summary

Surface air temperature in central Japan was predicted from the temperature recordings from sensors in the Automated Meteorological Data Acquisition System (AMeDAS), using seven different procedures: the usual simple and universal kriging and cokriging estimators, the traditional regression analysis and the inverse distance weighted method. The cokriging estimator integrated digital elevation data as well as the air temperature readings. The performance of the procedures was evaluated and compared using cross-validation.

The kriging estimator provided a better estimate than the traditional regression analysis that treated the data as spatially independent observations. The kriging estimate was also better than the inverse distance weighted method. Further improvement in the estimation accuracy was achieved by using cokriging procedures because of high correlation of air temperature with elevation. The accuracy of spatial prediction decreased due to nocturnal cooling in winter and daytime heating in summer. This decrease implies that a strong radiation balance at the surface, whether positive or negative, causes a relatively short-range variation in surface air temperature through the effects of local environments.