Kriged surface showing different scale

[MasoodShaikh]

I have data points (points shapefile) on the proportion of a certain disease, with values ranging from 0 to 1. After creating a kriged surface, the proportions ranged from 0.043 to 0.756. I need to understand why this discrepancy occurred. 

For reporting these findings in a scientific journal, would it be appropriate to use a scale ranging from 0 to 1, with the lowest class being 0.00 to 0.10 and the highest class being 0.70 to 1.00? Additionally, are there any references that explain why a kriged surface can have a different scale and best practices for using them in the map legend?

Thank you!

[EricKrause]

Hi @MasoodShaikh,

Kriging is an "inexact" interpolation method, meaning that the prediction surface does not pass perfectly through the values of the input points and instead has a tendency to smooth predictions (meaning that the range of predictions is usually more narrow than the range of the original data values).  The weaker the autocorrelation and the more noisy the data, the more it tends to smooth.  For your data, it's likely that you have many locations where high values are very close to low values, and the kriging surface effectively smooths over the highs and lows.

There are a couple things you can do.  First, you can use an interpolation method like Radial Basis Functions (aka splines) or Inverse Distance Weighting that will always honor the range of the input data values.  

Second, you can disable the Nugget effect of kriging, which will force it to honor the input data range.  You can do this on the semivariogram page of the Geostatistical Wizard by changing the “Model Nugget” option to “Disable”.  However, not using a nugget effect can often create strange artifacts in the output, so it is generally not recommended.  If you do this, pay close attention to strange behavior in the resulting surface.

-Eric