WRF Segmentation fault when reached radiation time step

[FedeJack]

WRF-4.5
I'm simulating three domains as set in the namelist.input1, for a 700 mm rainfall event in one day, rainfall starting at 9 am. Initial data is from ERA5. The simulation only runs successfully if I simulate the first two domains, so I used ndown to simulate domain d03 separately.
I ran wrf with max_dom = 2 and namelist.input1, then ndown.exe with namelist.input2.
After renaming the output files, I ran wrf.exe with only the d03 domain with the parameters in namelist.input2 moved to column 1. The simulation crashes with a segmentation fault when I reach the radiation calculation time step, even if I set it to 1, 3, or 9.
I've now tried with the default namelist.input, attached here as namelist.input3, but I get the same behavior (rsl.zip).
Thanks in advance.

 

[FedeJack]

Update: I try add in namelist3.input epssm = 0.2 and time_step = 1. This remove the CFL errors from all rsl.error* file but the segmentation fault when reached the radiation time step remains.

 

[kwerner]

Hi,
Apologies for the long delay in response while our team tended to time-sensitive obligations. Thank you for your patience.

Because the size of your domains are nearly identical between domains, using the ndown option to run domain 3 will not be beneficial. This is only helpful when the sizes vary so much that it becomes difficult to use the number of processors necessary to run the larger domain because that number is too large for the number allowed to run the smaller domain. So let's reduce part of the complication of this issue by going back to running all 3 domains simultaneously.

Can you try to run that again, decreasing the time_step and adding epssm to the namelist? You may also need to set smooth_cg_topo=.true. to the &domains section prior to running real.exe. Also, the setting for radt should be the same for each domain. So if you set it to 9 for d01, it should be set to 9 for d02 and d03, as well. If you try that and it fails again, please attach the updated namelist, along with the rsl* files again. Thanks!

 

[FedeJack]

When I set smooth_cg_topo=.true and run real.exe I get "found no soil elevation / terrain / topography data in metgrid files", see rsl.zip. I check with ncdump the met_em file and I find SOILHGT.

 

[kwerner]

The fatal error listed is :

-------------- FATAL CALLED ---------------
FATAL CALLED FROM FILE:  <stdin>  LINE:     813
            If using ERA5 data, possibly need to add more time invariant fields
-------------------------------------------

When you processed the ERA5 data, did you include an invariant file? See How to Process ERA5 Data for instructions on doing that.

 

[FedeJack]

I followed the instructions and now I'm wondering why everything worked without ERA5_INVARIANT with the smooth_cg_topo=.false option. But other than that, after run geogrid and ungrib, metgrid fails with:

Getting input fields from ERA5_INVARIANT
Processing LANDSEA at level 200100.000000.
Processing SOILHGT at level 200100.000000.
Preparing to process output time 2021-10-03_00
Getting input fields from FILE
ERROR: Cannot combine time-independent data with time-dependent data for field LANDSEA.mask

I tried both Vtable.ECMWF and Vtable.ERA-interim.pl (which I used before), with same result.
I have downloaded ERA5 data with the script DownloadERA5towrf.py

 

[FedeJack]

Is it correct to remove the LANDSEA row from the Vtable used by ungrib?

 

[kwerner]

I'm not sure if you should need to remove LANDSEA. Did you get these ERA5 files from Copernicus? We've seen some similar problems with files from that site. If that's what you're using, can you try to use the ERA5 files from RDA/CISL? Those files are in netCDF format, so you'll need to use a conversion script for them. See Converting ERA5 Model-level Data from netCDF to WPS Intermediate Format.

 

[zhanyx]

I'm not sure if you should need to remove LANDSEA. Did you get these ERA5 files from Copernicus? We've seen some similar problems with files from that site. If that's what you're using, can you try to use the ERA5 files from RDA/CISL? Those files are in netCDF format, so you'll need to use a conversion script for them. See Converting ERA5 Model-level Data from netCDF to WPS Intermediate Format.

Hello, I am also encountering a similar issue while using WRF version 4.5 and WPS with ERA5 data obtained from the CDS. I would greatly appreciate your advice on the following questions:

1. Could the data source website potentially be the cause of the problem, possibly due to format differences?
2. I have traditionally used ungrib.exe to process ERA5 data. Is era5_to_int intended to replace ungrib.exe, and is using this alternative tool necessary?
3. Are the files generated by era5_to_int directly compatible with metgrid.exe, and what specific settings should be configured in the namelist.wps to ensure proper processing?

Thank you in advance for your guidance.

 

[kwerner]

Hi, and first, apologies for the delay while we've been short-staffed over the past few weeks.

Could the data source website potentially be the cause of the problem, possibly due to format differences?

Without having any knowledge of the way the websites are built and download/store/process data, I'm not able to comment on whether this could be the case.

I have traditionally used ungrib.exe to process ERA5 data. Is era5_to_int intended to replace ungrib.exe, and is using this alternative tool necessary?

If the data you are using are not available in GRIB format (and are in netCDF format), then yes, this tool is necessary. It would also replace the ungrib process, so you would skip it and go straight to metgrid.

Are the files generated by era5_to_int directly compatible with metgrid.exe, and what specific settings should be configured in the namelist.wps to ensure proper processing?

Yes, the files are in intermediate format, which is the format of the files after they go through ungrib and are ready for the metgrid process. There are no special settings needed. Just run metgrid.exe.