Re: I: [ferret_users] Question about curv_to_rect_map

[Ansley Manke <ansley.b.manke@xxxxxxxx>]

Hi -
The averages might not be expected to be the same.  A true area-weighted 
average should be, but these are not both area-weighted averages.  The 
the average on the rectangular grid is area-weighted.  But the average 
you are doing of the original variable is not weighted by the area of 
the curvilinear grid cells but by the area of the index grid on which it 
is defined.

When you regrid to the rectilinear grid and then back again, that should 
be a better test.  I don't have an explanation for the consistent bias 
in a single direction.

We are working with the group which is developing libcf 
http://www.unidata.ucar.edu/software/libcf/, which will include a faster 
algorithm for regridding between curvilinear and rectilinear grids.  
We'll include tests of this kind as we work with that new algorithm.

Ansley

On 9/19/2011 3:16 AM, markus.mingel@xxxxxxxxx wrote:
> Dear All,
> I did one more test.
> I regridded the data from the new regular grid to the original curvilinear
> grid using the ferret external function rect_to_curv to check if I were able to
> get the same results of the original curvilinear grid, but as you can see from
> the figure results are differents (green should be equal to black).
>
> Is there a sound explanation?
>
> Thanks for the help,
>
> Best Wishes
> Markus
>
> > ----Messaggio originale----
> > Da: markus.mingel@xxxxxxxxx
> > Data: 16/09/2011 15.17
> > A:<ferret_users@xxxxxxxx>
> > Ogg: [ferret_users] Question about curv_to_rect_map
> >
> > Dear Ferretters,
> >
> > I'm using the function curv_to_rect_map to regrid a rotated pole grid to a
> > regular grid.
> > I don't have any problem running the script, but plotting the results after
> > the regridding I get a difference in global mean SST of about 1K.
> >
> > Here it is my script:
> >
> > define axis/x=0.5:359.5:1/modulo/units=degrees xax
> > define axis/y=-89.5:89.5:1/units=degrees yax
> > let dummy = y[gy=yax] + x[gx=xax]
> >
> > let lonin = LON[d=1]
> > let latin = LAT[d=1]
> >
> > let map  = curv_to_rect_map(lonin,latin,dummy,2)
> >
> > save/clobber/file=curv_map.nc map
> > CANCEL VARIABLE map
> > USE curv_map.nc
> >
> > DEFINE VARIABLE/title="NEW SST "     new_sst  = curv_to_rect(sst[d=1],map
> > [d=2])
> >
> > plot sst[i=@ave,j=@ave],new_sst[i=@ave,j=@ave]
> >
> > Obviously, after the regridding I don't expect to get 2 overlapped lines,
> but
> > I was surprised for a such difference. Do you have experience with that? Is
> it
> > maybe related to the average function in the curvilinear domains?
> >
> > Thanks for the info
> >
> > Best wishes
> > Markus