median polish in GRASS?

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

median polish in GRASS?

James Darrell McCauley

I have written code for a median polish algorithm that I will be
incorporating into GRASS. Median polish is a way of detrending
punctual data that separates large-scale and small-scale variation.
It requires a grid along with punctual data---each point is associated
with a node in the grid (this allows the algorithm to be used on
non-gridded data).  The residuals of this algorithm are suitable
for input to an ordinary kriging program.

For those GRASS users who know what this is and may have some
use for this feature in GRASS, tell me:

1. What would be the best way to store the output?
   (residuals would be stored as points-in-vect or sites, but
   what about {all}, {row}, and {column} effects?)
2. What type of post-polish analysis capabilities should such
   a GRASS tool have?  Cressie (1991) says something about checking
   for a cross-product trend in the residuals, but this doesn't have
   much meaning for me (programmer != statistician :-). Can someone
   enlighten me?
3. What about surface interpolation/extrapolation? Cressie (1991)
   gives equations for these that I could probably use to create
   a raster output map.  Would this be worth the extra effort?

If anyone wants to have any input, please email me (not the lists).
I plan to finish up this little project soon (end of the week) before
classes (and normal life) returns.

--Darrell McCauley ([hidden email])

Ref: @Book{ cressie91,
  author = "Noel A. C. Cressie",
  title = "Statistics for Spatial Data",
  publisher = "John Wiley \& Sons",
  year = "1991",
  series = "Wiley Series in Probability and Mathematical Statistics",
  address = "New York, NY"
}