I'm using Arcmap 10 and I have scattered points (denser in an area). Each point represents a thickness value of a simulated avalanche deposit. I want to get a raster from this points which I will transform in polygons (raster-to-polygons). Which is the best interpolation tool for this operation?

I tried them and I saw the "kriging" tool is the one which gives me the best "overlapping"* (i.e., the values of the points and the resulting raster coincide). Is this an appropriate way of establishing which tool is the one that suits better to my case?

Below, there is the image of the points superimposed on the raster obtained from them with the kriging interpolation.

enter image description here

  • 1
    Thickness of what, a geological layer? Maybe by "overlapping" you mean that the values of the points and the resulting raster coincide? If so, you maybe wish some exact method.
    – nadya
    Jul 31 '13 at 23:34
  • Thickness of a simulated avalanche deposit. Yes by "overlapping" I meant the values coincide. Do you think kriging could be the appropriate interpolation tool? Aug 1 '13 at 5:22
  • 2
    Kriging can be ok (if you wish it exact so without nugget) if you don't just click-click it with the default parameters, read here gis.stackexchange.com/questions/50584/…
    – nadya
    Aug 1 '13 at 16:41
  • Sorry, what do you mean with "nugget"? And which are the default parameters? I went through that post but it's quite a hard topic to me. Doesn't the observation of the picture in my question allow me to understand if the krigin is the interpolation tool I'm looking for? Aug 1 '13 at 20:36

You have a lot of points, so Kriging is probably the best way to go. (Geostats, like all stats need a decent sample size.) You can find a description of the "nugget" at:


However, if you are not familiar with kriging and geostatistics, you may want to use another exact interpolator (a tool that will preserve the values at your points), like spline or IDW. Be aware that tightly spaced points (which you have) with very different values (which doesn't appear to be the case) can give you a pretty funky spline. In this case, IDW is probably the easiest to use, and should give you accurate results with these tightly spaced points. The main problem with IDW is that it doesn't look good; the result tends to look a bit like a golf ball when viewed in 3D.

  • 2
    +1 Good advice. We can turn the proverbial lemon of IDW into lemonade: we know IDW doesn't do a great job, but with lots of data it can do an OK job. The bad look of the resulting surface is a useful reminder of these limitations. IMHO it's far worse to use a bad interpolator that produces great-looking maps, because that can (will) deceive people into trusting the results more than they should. (Splines are among the worst culprits.) BTW, for data like these--lots of points and seemingly good spatial density--you should consider TINs to be among the best (non-kriging) options.
    – whuber
    Sep 18 '13 at 14:32
  • I agree with the implementation of a Kriging model being the best approach. However, in looking at the data, I would imagine that, based on variable clustered sampled densities, the semivariogram will exhibit a "hole effect". This will have to be accounted for before a valid semivariogram can be specified for a Kriging model. Sep 18 '13 at 17:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.