Persistent crash in WRF (v4.5) that occurs after the 3rd model day [SIGTERM Error]

[Climate]

Dear WRF Community,

I’m currently troubleshooting a persistent crash in WRF v4.5 that consistently occurs after the third model day, with the following error:

forrtl: error (78): process killed (SIGTERM)

I'm running at 15 km resolution, using Noah-MP (sf_surface_physics=4) and the following physics setup:

&physics
mp_physics = 8, ! Thompson microphysics
ra_lw_physics = 4, ! RRTMG longwave
ra_sw_physics = 4, ! RRTMG shortwave
radt = 10, ! Radiation timestep (minutes)
sf_sfclay_physics = 1, ! MM5 Monin-Obukhov
sf_surface_physics = 4, ! Noah-MP
bl_pbl_physics = 1, ! YSU PBL
bldt = 0,
cu_physics = 0, ! No cumulus
cudt = 0,

What I’ve Tried and Observed:​

• The same configuration works perfectly in WRF v3.9. I’ve reused the same namelist.input and namelist.wps from that version.
• There’s no evidence of numerical instability in SLURM logs or standard output.
• The crash only occurs when the domain includes land. The run completes successfully over ocean-only domains.
• If I disable pressure level output (auxhist23), the simulation finishes without crashing, which makes me suspect an interaction with the land surface model or diagnostic output.

Other Notes:​

• I’ve confirmed that static fields (e.g., HGT_M, LANDUSEF, SOILTYPE) are present and appear normal in the met_em files.
• I'm using GHG_INPUT = 0, as required in newer versions.
• I’ve checked the rsl.error.* and rsl.out.* files, but no clear errors are reported before the crash (Attachment).

Given that the crash appears after ~3 days and is tied to land interaction or diagnostic output, I’m wondering:

• Has anyone experienced a similar issue with auxhist23 + Noah-MP in WRF v4.5?
• Could there be a subtle compatibility issue in WRF v4.5's land-surface handling or pressure level diagnostics?

Any advice or suggestions would be greatly appreciated.
Thank you!

 

[Ming Chen]

The section of &physics looks fine to me.

Can you upload the entire namelist.input for me to take a look? Also, what data did you use to drive this case?

Did you run WPSv4.5 to process data for this case?