I have a large set of polygons on a map, some of which contain data on 2 count variables, say ($z_{1}$ and $z_{2}$) that are correlated. In fact, $z_{2}$ most likely causes $z_{1}$ without the possibility of reverse causation. I need to interpolate $z_{1}$ to polygons with missing values, possibly with the help of $z_{2}$ within a co-kriging/co-located kriging setting. There is a couple of important aspects to the data:

  1. Both $z_{1}$ and $z_{2}$ oberved data measurements have standard errors that are independent of each other.
  2. There are also 4 types of polygons in my set, such that some have data on both $z_{1}$ and $z_{2}$, some have one or the other, and some have neither.


  1. How can I incorporate the measurement error from the input data to interpolate $z_{1}$, preferably with the help of $z_{2}$?

  2. Subsequently, is there a trade-off in terms of the accuracy of the resultant values to using co-kriging to interpolate $z_{1}$ to polygons that contain neither $z_{1}$ nor $z_{2}$?

Any advice, applied resources, code samples, or papers would be welcome.

I am aware of the issues surrounding Area-to-Area and Area-to-Point kriging, so let's assume these considerations are not part of this question. In full disclosure, I am piggybacking off this and this posts that provided some insight, but left me curious for more. Also, I am using R, so any advice on a package(s) that may be useful like DiceKriging is appreciated.

  • 1
    $\begingroup$ The techniques of geoRglm are the appropriate ones, but I don't know whether the software has implemented cokriging. $\endgroup$
    – whuber
    Commented Jun 21, 2020 at 13:44


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.