Key points are not available for this paper at this time.
Natural Resources Canada, Canadian Forest Service, and their partners have developed customized spatial spline models and gridded datasets for North America for a wide variety of variables, time steps, and spatial resolutions. The initial motivation in developing the models was to address forestry-related issues, however, many agencies and researchers have since used them in a variety of applications. The parameters for the basic model, along with the amount of data smoothing, are usually estimated by minimizing a diagnostic called the generalized cross validation (GCV). Average withheld error estimates for temperature and precipitation models across spatially representative locations at the monthly normal, historical monthly, and historical daily time steps, show that errors associated with the normal surfaces are small, reflecting the greater spatial coherence of monthly normals.
McKenney et al. (Mon,) studied this question.