Re: Natural neighbor interpolation/extrapolation

From: fred (fred@XXXXXX)
Date: Sat Sep 07 2002 - 17:16:40 MDT


  Patrick,

  Natgrid does not have a feature to do what you want. The only
  parameter that controls extrapolation is "ext" which controls
  whether extrapolation is allowed outside the convex hull (such
  extrapolation is allowed in the default environment). As you
  probably know you are on shaky ground when you extrapolate.
  Estimating how a function will behave when the only knowledge
  is in one direction works well only when the function is simple
  and consistent.

  The algorithm that Natgrid uses to extrapolate is as follows.
  It determines three points in the function domain space such that
  the triangle that connects these three points contains all points
  in the domain of the input function (these are called pseudo-points);
  it then calculates a plane that is a least-squares fit to the input
  data values; it then assigns function values for the pseudo-points
  such that these values lie on the least-squares fitted plane; it
  then considers the pseudo-points and function values as if they were
  in the original dataset. In effect extrapolated values asymptote to
  the least-squares fitted plane.

  Fred Clare

  ---------------------------------------------
  Fred Clare fred@ucar.edu 303-497-1284
  Visualization & Enabling Technologies Section
  NCAR Scientific Computing Division
  ---------------------------------------------

> Is it possible to use the extrapolation mode while telling
> it (the natural neighbor interpolation code) that for all extrapolated point
> which are beyond a certain range (say 50 meters) of the convex hul (the
> polygon created by the data I'm using as input for the natural neighbor
> operation), no extrapolation should be done, the default "no data" value
> should be used for those points?
_______________________________________________
ncarg-talk mailing list
ncarg-talk@ucar.edu
http://mailman.ucar.edu/mailman/listinfo/ncarg-talk



This archive was generated by hypermail 2b29 : Sun Sep 08 2002 - 21:30:25 MDT