NetCDF: Invalid dimension ID or name

pesieber · Apr 20, 2023

Dear CESM support,

I am trying to run regional cases at 0.5x0.5° and 0.1x0.1° resolution, which I built successfully using

./create_newcase --case $CASEDIR --compset I2000Clm50SpGs --res $CLM_USRDAT--mach $pizdaint --compiler gnu --driver mct --project sm61 --run-unsupported
./xmlchange CLM_USRDAT_NAME=$GRIDNAME
./xmlchange LND_DOMAIN_PATH=$REGDOMAIN_PATH,ATM_DOMAIN_PATH=$REGDOMAIN_PATH
./xmlchange LND_DOMAIN_FILE=$REGDOMAIN_FILE,ATM_DOMAIN_FILE=$REGDOMAIN_FILE

My regional domain over Europe goes over the zero longitude line and I changed the "xc" coordinate in the domain file such that longitudes are given in degrees east (i.e. positive values only), using: ncap2 -O -s 'where(xc<0) xc=360+xc' infile outfile

In my REGDOMAIN_FILE for 0.5x0.5°, lons are 300.75 to 84.25, and for 0.1x0.1° lons are 323.05 to 56.15.
I did not change the surface dataset because it has an index coordinate and I believe LONGXY can be in -180/180 degrees format without problems.

The run completes successfully, and writes reasonable output for the right domain. However, in the cesm log I get "NetCDF: Invalid dimension ID or name". Is this a cause for concern? There is no other error in the cesm log, lnd log or atm log (see attached). Do I forget something, e.g. do I need to supply mapping files to any of the components?
Actually, with an independent build with the same settings I completed one run without the NetCDF message, but I cannot reproduce it.

Thanks in advance for your help,
Petra

pesieber · Apr 20, 2023

I forgot to mention the important difference to the build without the NetCDF message: everything is the same, except that for 0.5x0.5° resolution I use hcru_hcru in place of CLM_USRDAT. I want to "switch" to CLM_USRDAT so I can use the same workflow for both resolutions (and I am not aware of a defined grid for 0.1x0.1°).

oleson · Apr 20, 2023

I'm not sure if it is a problem. It might be worth trying to build and run in DEBUG mode to see if the model stops and you get a traceback.

pesieber · Apr 20, 2023

From the output I see that it completed all time steps that I tested (right now only 24h with hourly output). Do you think I could see more in DEBUG mode even if the model does not fail?

It seems to be related to CLM_USRDAT (since I don't get it with hcru_hcru and the same 0.5° domain file). Is there something specific the model does/reads with this grid?

I guess for CLM with my compset it does not matter at which resolution it is compiled, because there are no lateral exchanges, so possibly I could compile with hcru_hcru and provide the 0.1° domain file?

oleson · Apr 20, 2023

I don't think you'll see more in DEBUG mode if the model doesn't fail. I'm just thinking that building/running in DEBUG might make the model fail and then give you a traceback if the NetCDF error is indeed a real problem.
I don't think it necessarily has anything to do with the grid itself. It's possible that there is a variable or dimension missing in one of the domain or surface dataset files, etc. that is not critical.

oleson · Apr 20, 2023

I do see this post which may indicate that it's not a problem: