Bug Report: Grid-Stat crashes with UGRID-compliant MPAS data (NetCDF::NcBadId exception)

[USAF-SC-16WS]

Hello MET Support Team,

I am currently building an MPAS unstructured grid data verification tool integrating MET and UGRID, and I have encountered a crash in Grid-Stat when attempting to read my data.

Below is a detailed bug report. Please let me know if there is a known workaround or if you need me to provide the sample files.

Summary
Grid-Stat (v12.2.1) crashes with a netCDF::exceptions::NcBadId exception when attempting to read UGRID-compliant NetCDF files containing MPAS unstructured mesh data. The file is valid CF-1.8/UGRID-1.0 compliant NetCDF that can be read by standard tools (ncdump, Python netCDF4), but MET's UGRID reader fails.

Environment

• MET Version: 12.2.1 (Docker image: dtcenter/met:12.2.1)
• Platform: Linux (Docker container)
• Command:
• grid_stat /output/ugrid/mpas_ugrid.nc /data/gdas.t00z.atmanl.nc /output/ugrid/GridStatConfig -outdir /output/ugrid

Error Message
terminate called after throwing an instance of 'netCDF::exceptions::NcBadId'
what(): NetCDF: Not a valid ID file: ncDim.cpp line:80
FATAL: Received Signal Abort. Exiting 6

Steps to Reproduce

1. Create a UGRID-compliant NetCDF file from MPAS data with the following structure:
   * Mesh topology variable with cf_role = "mesh_topology"
   * Face-node connectivity (verticesOnCell)
   * Face-face connectivity (cellsOnCell)
   * Coordinate variables (latCell, lonCell) in radians
   * Data variable (t2m) with dimensions (Time, nCells)
   * Time coordinate variable with proper units

2. Create a Grid-Stat config file with:
   fcst = {
   file_type = NETCDF_NCCF;
   field = [
   {
   name = "t2m";
   level = [ "(,)" ];
   }
   ];
   }

3. Run Grid-Stat with the UGRID file as forecast input.

4. Grid-Stat crashes with the NetCDF exception.

Expected vs. Actual Behavior

• Expected: Grid-Stat should read the UGRID-compliant file and perform verification statistics on the unstructured mesh data.
• Actual: Grid-Stat crashes with a low-level NetCDF C++ exception (NcBadId) in ncDim.cpp line 80. This suggests the UGRID reader is attempting to access an invalid NetCDF dimension or variable ID.

Additional Context & Validation

• The file is valid and can be read by ncdump, the Python netCDF4 library, and standard NetCDF tools. It follows CF-1.8 and UGRID-1.0 conventions, has proper mesh topology metadata, valid coordinate variables, and a proper time dimension.
• This is for MPAS (Model for Prediction Across Scales) verification against GDAS analysis.
• MPAS uses an unstructured Voronoi mesh (370,681 cells). UGRID is the recommended approach for this in CF conventions.
• Note: MET's regrid_data_plane also fails with "no grid defined" when trying to read raw MPAS files directly.

Questions for the Team

1. Is UGRID support for unstructured meshes fully implemented in MET 12.2.1?

2. Are there known limitations with UGRID mesh sizes (370k+ cells)?

3. Is there a specific UGRID file format or metadata that MET requires beyond the UGRID-1.0 specification?

4. Would MET 13.0.0 (currently in beta) have better UGRID support for this?

5. Are there example UGRID files that successfully work with MET Grid-Stat that I could reference?

I am happy to provide the sample UGRID NetCDF file, my complete Grid-Stat config file, or full error log outputs upon request.

Thank you for your time and assistance!

Best regards,
Sierra Cox
Software Developer
WXS - 16th Weather Squadron
USAF

[hertneky]

Hi Sierra,

It looks like you are having issues running GridStat with your MPAS UGRID file against a gridded GDAS analysis. While the capability was implemented with v12.0.0+, there are certainly some limitations. For one, GridStat cannot verify directly on the UGRID and needs to be regrid to a structured grid. You can do this in the GridStat config with the regrid option. You can either regrid to OBS or other structured grids. At this moment, the only regrid method for unstructured grids is NEAREST, however, bilinear interpolation is slated for the next release (see dtcenter/MET#3334). Another note on regridding, is that, while auto-regridding for UGRIDS can be done within GridStat, the RegridDataPlane tool does not currently support UGRIDs.

Here is an example GridStat config for MPAS that is used by the METplus unit-tests. You can find the metadata requirements for UGRIDs here.

If you are still having troubles, please send your NetCDF file and GridStat config file to us. You can do so with these instructions. If the options in those instructions don't work for you, I think we have had the AF send us data through dropbox as well.

Thanks and let me know how it goes!

[hertneky]

Hi Sierra,

I am checking in to see if this problem is still persisting for you, or if you have figured out a solution.

[USAF-SC-16WS]

So sorry for the very delayed response, yes we found a way forward by re-gridding the unstructured data we were able to get an output from grid-stat then plot with cartopy and matplotlib. Thank you for your assistance and again sorry for the delayed response.