Re: random point interpolation

From: Dennis Shea <shea_at_nyahnyahspammersnyahnyah>
Date: Fri, 5 May 2006 11:09:45 -0600 (MDT)

>I cannot find an example of this in either the talk archives or the
>docs, but hopefully somebody out there has figured out an efficient way
>to do this.
>I have a coarse-resolution grid (from WRF) with the associated 2D lat
>and long arrays. The lat and long are of course not monotonic. I would
>like to interpolate to another WRF-like grid, with similar traits. The
>output grid is huge so I don't want to loop through individual points in
>my script. Does any single function handle this?
>For example:
>inLon = (/ny,nx/)
>inLat = (/ny,nx/)
>inVar = (/ny,ny/)
>outLon = (/NY,NX/)
>outLat = (/NY,NX/)
>outVar = (/NY,NX/)
>I want take inVar to the grid defined by outLon and outLat.
>Again, none of these are monotic if unrolled to vectors.


In an offline email, Josh clarified that NCL's
nearest neighbor interpolators

were not appropriate for his needs.

Further, the data are not randomly located.
The data are on logically rectangular grids.

I suggested a non-NCL package:

    The Spherical Coordinate Remapping and
    Interpolation Package (SCRIP)
Given input grid information contained on a netCDF file,
SCRIP uses fortran-90 code to generate an array of
weights that are written to a output netCDF file.
The weights on the output netCDF file can be
read via NCL. They can be used to very efficiently
interpolate from one grid to another. The NCL function 'pop_remap'

can then be used to apply the weights.

FYI: This is the approach used by the Community Climate
Systems Model [CCSM] to communicate 'information' to/from
the various component models which are all on different
logically rectangular grids.


ncl-talk mailing list
Received on Fri May 05 2006 - 11:09:45 MDT

This archive was generated by hypermail 2.2.0 : Fri May 05 2006 - 15:37:41 MDT