Predicting in advance which is the best interpolation method for my data

Question

I need to interpolate in ArcGIS 10.1 soil data (organic matter content, nutrient content, pH, etc.) from a set of soil sampling points to the surface which this set of points lays on.

I am considering Kriging or Inverse Distance Weighted.

I guess that using the Geostatistical wizard (in the Geostatistical Analyst toolbar) is the best option since the wizard has “Optimize the entire model” and “Optimize power value” buttons for respectively both interpolation methods. But I have seen that for certain attributes or values to interpolate, it can be obtained a lower RMS by Kriging than by IDW, and on the contrary for other attributes.

How to know in advance which method gives better interpolation result (lower RMS –or any other quality indicator-)? Or, on the contrary, is it necessary to achieve it by trying to change as many different parameters of both methods as possible?

Answer 1

By design, kriging is the best linear interpolation method for a single input variable, thus it is a better method than IDW (which is also a linear interpolation method for a single input variable). Indeed, kriging minimize the errors of prediction.

The "problem" with kriging is that it is more complex than IDW, so it takes more time (and more skills) to build a good kriging model than to find the best IDW model(which is possible by "brute force"). However, if you take time to look at your data, you do not need a cross validation of all parameter to run your kriging. You need to select the best model according to the semi variogram (there are different types of kriging and different advanced parameters, but semi-variogram is the main one).