best sampling design to perform kriging given old measurement in the same area

Question

This question should better fit Crossvalidated than stackoverflow, but my questions on kriging never find an answer there, while they do in here, so please do not move the question.

In a project we sampled the DVB-T field and I made some kriging interpolation. A new measurement campaign is in the air is there a way to know, given the old measurement what is the best sampling design and how many measurements should be done?

I checked on the Cressie, that sent me to a ton of other articles and I looked a lot in Google, but it seems I cannot find the right reference.

I do not want an iterative method, that is the main deal.

Any type of reference is welcome.

Answer 1

At the end I think I will opt for spatial simulated annealing, I am still open to other options, though. Please let me know if you have any.

Answer 2

It is surely a bit late but here is an answer to you problem...

If you just want to improve you model, you can put the new points where the variance of your kriging predictor is the highest.

If you want to optimize the field you are interpolating (i.e. find its minimum or maximum), then you can use the expected improvement which is a criterion which tells you where to sample the next batch of points. See, for example: https://lirias.kuleuven.be/bitstream/123456789/310611/2/JGO_2012.pdf

Another approach could be to sample the bnew batch of points "more indepently" from the kriging interpolator you have computed. You can, for example, choose the next point so that the sample containing the previous sample points + the new ones minimizes a space-filling criterion like maximin criterion or a discrepancy criterion. For details about these types of criteria, see, for example, http://arxiv.org/abs/1307.6835 and the references inside.