Can locations used for ordinary kriging be weighted differently?

Question

Hi Everyone!

I am using ArcGIS Pro 2.4.0 with full licensing.

I have a dataset that I would like to interpolate using Ordinary Kriging. This dataset uses the average of two measurements. However, I have a sizable amount of locations with only have one measurement. Is there a way that I can "weight" the locations that only have one measurement so that the Kriging analysis does not treat these two different measurements as equals?

For example, is there a way that I could make a column and add a "2" if the location has two samples averaged or a "1" if the location only has one sample, then have the Kriging factor in that difference?

Thank you!

DanPatterson · Answer

That is right  Essentially they are  removed. (values averaged, first or last used... depends on the software)

There is no priority weighting however.

Eric6346 · Answer

If you are using Geostatistical Analyst and the Geostatistical Wizard, you can accomplish this by creating duplicate records in your feature class. Any measurement that is the average of two values should be coded as two coincident points.

In the Geostatistical Wizard, these coincident points will be identified, and you'll be asked how to reconcile them.  The default Mean option accomplishes what you want.  Internally, it averages the values of coincident points but keeps an internal weight that is used when estimating the semivariogram.  In the search neighborhood, the coincident points are treated like a single location, so you will not fill up the search neighborhood with multiple points at the same location.

DanPatterson · Answer

Duplicate locations get dealt with.  Essentially they are  removed. (values averaged, first or last used... depends on the software)

Nudging a duplicate a teeny bit so you can use its value affects the interpolation for all the surrounding space (generally not a good idea)

Replicating a measure at the same location doesn't do much unless you do it at all previous locations (this is data replication testing)

Measuring in a new location, perhaps in between previous measures may be useful, but you shouldn't do it selectively, it defeats a useful sampling strategy.

Given these considerations, why would you want to weight a location's value as being more important than surrounding values?