Re: questions about interpolation by NCL

From: Dennis Shea <shea_at_nyahnyahspammersnyahnyah>
Date: Thu, 7 Dec 2006 07:18:49 -0700 (MST)

> Indeed, the spectral harmonic method spreads the whole
> domain while the bilinear mothed is more local. I've
> tried to use linint2_Wrap function to regrid the
> topography and its flag(ocean=0,land=1), the results
> are relatively rational and acceptable. However, I'm
> now puzzled that for what kind of occasions we should
> use the f2fsh_Wrap/f2gsh_Wrap and linint2_Wrap, and
> which is better of the two kinds of regrid functions?
It depends upon the application. :-)

As a general rule, I would recommend the bilinear
interpolation [linint2 , linint2_Wrap]. You will never
get an interpolated value that exceeds the field min/max.
Also, this function handles missing values.

Most frequently, the spectral regridding functions are
used when users want to compare

     [a] observational grids on (say)a 2.5x2.5 grid
         to spectral model output fields. Use
         of "f2gsh(grid,outDims,twave)" creates
         a grid where the spectral truncation used
         is the same as the model.

     [b] different model outputs: g2gsh creates grids
         with similar spectral truncation.
The spectral methods are faster than the bilinear interpolation.
This is noticible when the grids are large and there are many
levels and time steps.

If the variable being interpolated is not bounded [eg, zero for 
precip] and is smooth then either method will provide
reasonable fields. 
ncl-talk mailing list
Received on Thu Dec 07 2006 - 07:18:49 MST

This archive was generated by hypermail 2.2.0 : Thu Dec 07 2006 - 16:52:37 MST