Specify constants in Universal Kriging Second Order Polynomial

[Lumera]

Hello lovely humans!

I am analyzing a data set of soil nitrogen levels and want to use Universal Kriging to create an interpolated prediction surface. The data is skewed to the left of a normal curve (no non-zero values, the study site is in an arid ecosystem so most values are relatively small). I am using a BoxCox transformation with a power of 0.5 to help account for the skewedness. The data is also not stationary as it has a strong seasonal influence, hence my choice of Universal Kriging. 

I have values for each data point for three separate years (in other words, there are three distinct values for every unique coordinate location sampled) and have been running Universal Kriging through the GeoStats Wizard and selected to use all data points as they represent unique nitrogen values (see screen shot). 

*If you have any feedback on the approach I am using I am open to learning, I have been muddling my way through this on my own and have much to learn 🙂

My main question comes in at this point of the analysis:

I have good reason to think that the season (which strongly correlates to precipitation in this study) with impact the soil nitrogen levels and therefore want to incorporate it into the polynomial being used. I also have two other factors (species and shrub presence) that may also be informative of nitrogen levels and would like to explore their impact on the model. However, I cannot figure out how to specify the constants in the polynomial equations in the Geostats Wizard (in other words, I would like to specify the data that the model uses to identify the constants in the polynomial equation). Is this an option in ArcGIS Pro?

I cannot in good faith use regression kriging as my data is not stationary (unless y'all have some input on how I can control for that). 

Any help is much appreciated!

[DanPatterson]

post moved to Geostatistical Analyst questions where you might get an answer rather than in the basic arcgis pro section of Community

[Lumera · Accepted solution]

So after lots of digging, I think the answer to my question is there is no direct way to define the constants. Arc runs all the math on its own and you can then interpret those results in light of what you know about your data set. If you are in a similar boat, it may be worth looking into EBK Regression kriging as it can handle non-normal and non-stationary data, though it has its limits.

[EricKrause]

Hi @Lumera,

As you figured out, the coordinate polynomials in Universal Kriging will be calculated and used automatically.  They cannot be manually overridden.

The closest equivalent is EBK Regression Prediction, where you can use a raster that will be used in a regression equation to determine the mean.  However, the regression coefficients are not configurable, so the mean surface will be in the form: Mean(s) = B0 + B1*RasterCellValue(s).  That's not exactly the same as defining the mean surface, but it's close.