joining two datasets into one

[JuanNogues]

I have two datasets that have the same information (point values of groundwater level). However one is of 2007 and the other of 2001. I would like to know if there is a method, tool or way to prove statistically that these datasets can be joined to be one or a way that I can prove that they can't be joined.

More specifically, I have 46 points in one data set that intorpolating gives me one map and another with 60 points that give me another map. The points are all in different locations. Ideally I would like to join the 106 points into one data set in order to create a more accurate interpolation. But to do that I would have to prove statistically that these points are from the same population. In common statistical analysis there are several tests that could be done. However with geostatistics I am not sure how to proceed.

Any help in the form of advice, pointing to a previous post, a guide or a solution would be greatly appreciated.

Thank you

[EricKrause]

My recommendation would be to not attempt to merge the two datasets.  Instead, perform cokriging.  Cokriging will use information from both datasets to make predictions, but it won't assume that the two datasets have the same statistical properties. 

Even if the two datasets do have identical statistical properties (implying you can safely merge them), you won't lose much by performing cokriging compared to merging the datasets and performing univariate kriging.  The only disadvantage is that you will have to estimate two semivariograms and one cross-covariance curve rather than just one semivariogram.

[JuanNogues]

Thank you Eric6346..... I will try this - though I have to read up on co-kriging to understand the assumptions that I have to make regarding a cross-covariance curve.  Any hints on where to find some quick reading on this topic ?

Thanks again.

My recommendation would be to not attempt to merge the two datasets.  Instead, perform cokriging.  Cokriging will use information from both datasets to make predictions, but it won't assume that the two datasets have the same statistical properties. 

Even if the two datasets do have identical statistical properties (implying you can safely merge them), you won't lose much by performing cokriging compared to merging the datasets and performing univariate kriging.  The only disadvantage is that you will have to estimate two semivariograms and one cross-covariance curve rather than just one semivariogram.

[EricKrause]

You'll fit the cross-covariance curve the same way you fit a semivariogram.  You'll see empirical covariances (blue crosses), and you need to manipulate the parameters (range, nugget, sill, lag size, etc) to get the curve to go through the crosses as best you can.  The first thing to try should be the "Optimize Model" button.  Hopefully that will be able to fit everything automatically, though you may want to do some manual tweaking from there.

Also, if you're doing external research, the cross-covariance curve is often called a "covariogram."