How to remap variables with "nvertices"dimension ?

[zzzeamon]

Hi developers,
I am a beginner with E3SM. I encountered an issue while remapping data. I followed the command below to remap sea ice output, but found that this command only works for variables with the 'ncell' dimension.

ncremap -m map_EC30to60E2r2_to_cmip6_180x360_nco.20221012.nc dat_src.nc dat_rgr.nc

Some variables ( uVelocityGeo,vVelocityGeo...) in the sea ice output have the 'nvertices' dimension and cannot be remapped using this command. Could you please advise on how to solve this?

can be remapped:
timeMonthly_avg_iceAreaCell
Size: 236853x1
Dimensions: nCells,Time
Datatype: single

cannot be remapped:
timeMonthly_avg_uVelocityGeo
Size: 482371x1
Dimensions: nVertices,Time
Datatype: single
Attributes:
units = 'm/s'
long_name = 'True eastwards ice velocity'

[xylar]

@zzzeamon, my hope is that @czender has an NCO solution for this.

If not, an alternative is the (poorly documented) pyremap python package (available via conda-forge). There are some examples, and the closest to what you want is:
https://github.com/MPAS-Dev/pyremap/blob/1.4.1/examples/make_mpas_to_lat_lon_mapping.py
Here is shorter suggestion:

#!/usr/bin/env python
import xarray

from pyremap import (
    LatLonGridDescriptor,
    MpasVertexMeshDescriptor, 
    Remapper,
)


method = 'bilinear'
date = '20250319'

in_descriptor = MpasVertexMeshDescriptor(fileName='dat_src.nc', meshName='EC30to60E2r2')

out_descriptor = LatLonGridDescriptor.read(fileName='dat_rgr.nc', meshName='cmip6_180x360')

map_filename = f'map_{in_descriptor.meshName}_to_{out_descriptor.meshName}_{method}.{date}.nc'

remapper = Remapper(in_descriptor, out_descriptor, map_filename)

remapper.esmf_build_map(method=method, mpi_tasks=1)

[zzzeamon]

Thank you very much for your detailed guidance, I successfully remap the vertex-field.

[xylar]

Glad that worked!

[czender]

NCO does not have a solution for this. Please try the solution above provided by @xylar.

One difficulty for NCO is that the user is attempting to regrid a field stored on cell vertices, not cell centers. Note that the cell vertices are present only in default MPAS restart files, not monthly history files. The vertice locations could be obtained from a restart file and then converted to their own grid for application of the usual algorithms. It looks like the MpasVertexMeshDescriptor method above uses MPAS secret sauce to infer vertices locations from fields available in the history files. NCO lacks this functionality. However I can put it on the TODO list if the method above does not work for you @zzzeamon.

[xylar]

Just to clarify, MpasVertexMeshDescriptor requires a file with a valid MPAS mesh such as a restart file, so it cannot infer vertex locations from a history file. The native data on ESGF should always have included such a restart file, I believe.

[zzzeamon]

Thank you! I appreciate your efforts making data processing so convenient.

[philipwjones]

If you're adventurous, you can create a mesh file for a vertex-centered mesh with the neighboring cells as cell corners. Then you can generate a map for that mesh to whatever destination you want. Since velocities are actually better interpolated without a conservative (area-based) remap, you could also use a vertex-centered mesh to do a nearest-neighbor distance weighted interpolation or other point-wise solution (see, for example the constrained optimization remaps by Bochev et al. https://doi.org/10.1016/j.jcp.2013.03.057 )

[xylar]

Bilinear interpolation using the MpasVertexMeshDescriptor should produce results that are pretty similar to this approach. But I guess the idea here was if you happened to know the locations of cell centers but not vertices?

[zzzeamon]

Thanks for your approach.