Multivariate data analysis: Iterative Universal Kriging


Here are posted the codes implementing Iterative Universal Kriging (IUK) as described by Sherwood (2000) (for homogeneous data) and Sherwood (2007) (for inhomogeneous data). It is designed for superior estimation of means, trends, or other signals in data from a set of fixed stations with incomplete sampling -- including station sets with heterogeneous periods of operation and reporting rates. See the aforementioned articles for a detailed discussion of the statistical approach and performance characteristics. This method has been applied to the radiosonde network by Sherwood (1999) and Sherwood et al. (2007).

The current version (2.21, 05/2007) of the method is equipped to quantify level shifts associated with artificial change-points. The times of these change points must be supplied by the user. If change point times are not known a priori, the user must detect them from the data using some method such as those tested by Sherwood (2007). Output from the first model fit can be useful for an improved detection, as shown in that study; the resulting change-point times can then be supplied for a second, more accurate run of the analysis. The software (a 1.5 MB tar file) is currently available only as IDL source code. V2.21 fixes a minor bug which was present in V2.2.

Download IUK version 2.21
It comes ready-to-run with a sample data set. Limited documentation is provided as a README and as comments in the top-level scripts.