Answers are grounded in this paper and cite their sources.
In a comparative experiment, the sequence of daily maximum and minimum temperatures for 1976 was interpolated over England and Wales to a resolution of 1 km using partial thin plate splines, ordinary kriging, trend surface, and an automatic inverse-distance-weighted method of interpolation. A "level playing field" for comparing the estimation accuracies was established through the incorporation of a consistent set of guiding variables in all interpolators. Once variables were included to guide the interpolators, differences in estimation accuracy among partial thin plate splines, ordinary kriging, and inverse distance weighting results were not significant although the performance of trend surface analysis was poorer. Best accuracies were achieved using partial thin plate splines, with jackknife cross-validation root-mean-square errors of 0.8°C for an annual series of daily maximum temperatures and 1.14°C for daily minimum temperatures. The results from this study suggest that sole reliance on the selection of guiding variables can be a less efficient means of achieving the required accuracies than the placing of greater reliance on empirical techniques of interpolation that can account for known autocorrelation in the temperature data. The use of guiding variables narrows the gap between performance of the different interpolation methods. In general, however, more sophisticated interpolators such as kriging and splining require fewer guiding covariates to achieve similar estimation accuracies. Day-to-day variability in the interpolation accuracies confirms the need for increased adaptability in the manner in which the guiding variables are incorporated in the interpolation process.
No takes yet. Share an insight, caveat, or question.
Jarvis et al. (2001) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: