Reproduciblity Issue with 'create_region' -- high-resolution (3km) meshes

[mdtoy65]

Hello,
Colleagues of mine used 'create_region' to cut out an elliptical domain from the 3km-uniform global grid 'static' file provided at MPAS-Atmosphere mesh downloads, i.e., x1.65536002.static.nc. The 'pts' file used was:
Type: ellipse
Point: 28.5, -67.0
Semi-major-axis: 1800000 # Meters
Semi-minor-axis: 4250000 # Meters
Orientation-angle: 5

I then needed to cut out the same domain from the 'x1.65536002.grid.nc' file, and attempted to do so with the same 'create_region' script and 'pts' file, however a slightly different grid resulted. The 'static.nc' file that was generated had 'nCells = 3014402'. The regional 'grid.nc' file ended up with 'nCells = 3014403', that is, it had one extra grid point. Looking at the differences between the 3km global static.nc and grid.nc files is that the 'latCell' and 'lonCell' variables are 'double' precision in the 'grid.nc' file and 'float' in 'static.nc'. I'm wondering if this is the cause of the mis-matched meshes, i.e., due to differing round-off errors.

-- Mike

 

[Ming Chen]

Hi Mike,
I saw the similar behavior before and I guess it is due to differing round-off errors, although I am not 100% sure.
However, I don't think this could be an issue that will affect MPAS run. Anyway only static.nc will be used when you run mpas_init_model and mpas_model.

 

[mdtoy65]

Ming,
This is an issue because we are already running on the mesh produced by the regional cut-out of 'static.nc'. However, I am introducing a new physics scheme that uses static topographic data by running 'init_atmosphere_model' to produce a static file that is read by 'atmosphere_model'. It's important that the meshes be identical. I am going to try substituting the 'double' lat/lon variables in 'grid.nc' with the same variables but with 'float' precision to see if this fixes the problem.
Thanks,
Mike

 

[Ming Chen]

Mike,
Thank you for the detailed description. Please keep me updated of your result after changing the datatype to 'float'. Experiencing the same issue before, I am really curious how to fix it. Thanks in advance.
Ming

 

[mdtoy65]

Ming,
The results of my changing the datatype of all the lat/lon variables (Cell, Edge and Vertex) in 'grid.nc' from 'double' to 'float' had the desired result -- the number of Cells, and the mesh itself, now matches that made by running 'create_region' on the 'static.nc' file.
Regards,
Mike

 

[Ming Chen]

Mike,
Many thanks for the update !