[MITgcm-support] netcdf with curvilinear grid

[Martin Losch]

Hi there,

I am using a curvilinear grid much like Dimitris Menemenlis' arctic  
configuration. It's based on a rotated grid where the north pole has  
been rotated to the equation, etc. I divide my domain into two tile  
for two processors:
snx=243
sny=85
nsx=1
nsy=1
npx=1
npy=2
so Nx=243, Ny=170

I have now managed to get the model to read the grid  
usingCurvilinearGrid=.true. (BTW, only the OLD_GRID_IO variant works,  
with the the tile00?.mitgcm, I run it all sorts of problems, but  
that's a different question), and everything seems to run OK. But the  
netcdf output is unexpected. The first tile comes out OK, but the  
second tile, which should be on top of the first one is flipped, it  
basically runs from j=170 at the bottom backwards to j=86 at the top.
Probably as a consequence, Baylor's fantastic script gluemnc does not  
work and it puts the two tiles next to each other (so that I get a  
domain Nx=2*243,Ny=85).
Matlab's rdmnc on the other hand seems to be doing the right thing  
and flips the 2nd tile back to normal.

What's going on?
Martin