s.sample

James Darrell McCauley mccauley at ecn.purdue.edu
Mon Apr 17 14:04:28 EDT 1995


Wayne Gibson (oregon at chaos.ats.orst.edu) writes on 17 April 1995:
>Can someone explain the advantages of using bilinear interpolation vs cubic  
>convolution in s.sample?
>
>The cell values of the raster data set that is being analyzed represent areal  
>averages.  From the GIS books that I've consulted, cubic convolution gives more  
>stable areal estimates.  But are there any specific cases where bilinear  
>interpolation would give better estimates?  (e.g. high gradient vs low gradient  
>areas)

[I'm the author of s.sample, so input from others would be welcome
here -jdm]

I suppose that one practical consideration is edge effects.  Cubic
convolution uses a 4x4 neighborhood whereas bilinear interpolation
uses a 2x2 neighborhood. Do you need to sample near edges?

Another (more important) consideration is the underlying model of your
data: does cubic convolution (higher order interpolation) make sense?
Are you making good assumptions regarding the continuity of your data?
Cubic convolution gives a much smoother estimate.

[As an aside, in addition to the three basic re-sampling techniques
available in s.sample, there are several models that people use for
resampling. See, e.g., Mitas & Mitasova's thin plate splines and
r.resample.tps.]

What do you intend to do with your resampled data?  Richards (1993)
recommends cubic convolution if the final product is to be used for
photointerpretation but cautions against it if you intend to classify
the re-sampled image. He cites the following in his discussion of
cubic convolution:
  T.G. Moik, 1980: Digitical Processing of Remotely Sensed Images, 
    Washington, NASA.
  S. Shlien, 1979: Geometric Correction, Registration and Resampling of
    Landsat Imagery. Canadian J. Remote Sensing, 5, 74-89.
  K. Simon, 1975: Digital Reconstruction and Resampling for Geometric
    Manipulation. Proc. Symp. on Machine Processing of Remotely Sensed
    Data, Purdue University, June 3-5.

If anyone is at all unsure of the math involved in s.sample, please
see:

ftp://pasture.ecn.purdue.edu/pub/mccauley/grass/tutorials/s.sample-tutorial.ps.gz

(the tutorial doesn't go into details of sampling theory--it only gives
the equations for what's going on under the hood in s.sample)

--Darrell

James Darrell McCauley, PhD        http://soils.ecn.purdue.edu/~mccauley/
Dept of Agricultural Engineering   mccauley at ecn.purdue.edu
Purdue University                  tel: 317.494.1198 fax: 317.496.1115

P.S. The ref I cited is:
@Book{ richards93,
  author = 	 "John A. Richards",
  title = 	 "Remote Sensing Digital Image Analysis",
  publisher = 	 "Springer-Verlag",
  year = 	 "1993",
  edition =      "2nd",
  address = 	 "Berlin"
}




More information about the grass-user mailing list