Thank you for your answer. It seems that there is a detail (or more) that I don't get; I develop more what I am doing below, with illustrations in the attached file, hoping that it will help to determine what I don't understand.
I have two options to perform an Ordinary Kriging interpolation:
A. The Geostatistical wizard: figures A.1 to A.9.
B. A tool from the toolbox Spatial Analyst: figures B.1 to B.2
When performing A.1 to A.9, I obtain a smooth output (A.9).
When performing B.1 to B.2, I obtain an output that shows some structure (B.2).
My conclusion, at first, was that option A (the wizard) generates more appropriate default interpolation parameters (sill, range, lags, ..) than option B (the tool from the Spatial Analyst toolbox).
As it seems that I can only use the tool from option B in my script (whose prototype is shown in B.3), my question was: is there a way to have the tool from B computing default parameters the way A (the wizard) does. Or is there a method of the Geoprocessor that would provide me with parameters similar to what the wizard generates (but it has to be the output of a method (a function), because it is not an operation that I can repeat by hand), so I can setup the tool form B with them.
Reading your answers, I thought at first that B (the tool from Spatial Analyst) could be using an XML file outputed by the wizard (with auto="true") and would then be computing default parameters the way A (the wizard) does. But I can see no such input in B (on screenshot B.1 or in B.3 for the method), and I am unable to find in A.1 to A.9 where it is possible for me to output an XML file with interpolation parameters.
I hope that this explanation shows you clearly what I don't understand.
There is an additional point that comes to my mind: is it possible that the apparent smoothness in A comes from the conversion to raster that is done in a way "as smooth as possible" while still being within the estimation of the error (?), whereas in B we just have the best estimate? With Kriging I can expect the interpolant to have a symmetrical pattern; the surface is moving towards the mean when getting far form data points, and the symmetry comes from the regularity of the grid. The reason I used earlier the word "artifacts" is that I was expecting smoother "ups and downs" I guess and I was surprised to see a more geometrical pattern. If this is correct, then I have no more question because it means that B (the one that I can use in my script) is a better interpolant for me than A (that is no more a best estimate).
Thank you and best regards,
Cedric