[SOLVED] Segfault for WRF V4.5 with the new hybrid 100-m MODIS-LCZ dataset using SLUCM urban physics

[winstonwu91]

Hi,

I came across segfault for WRF 4.5 using the new modis-lcz dataset under SLUCM urban physics. The segfault always happen at 1 of the 72 cores in Cheyenne. From the error log it seems that the segfault happens during avgflx at solve_em.F.

I cannot pinpoint how it happens without any information. When preparing the met_em files I switched to the GEOGRID.TBL.ARW_LCZ file.

Here is the rsl.error.0006 file that is giving me error, along with the rsl.out.0006 file (in rsl.out there I coded the output of ZDC, Z0C and ZA for debugging purposes). namelists are attached also. In addition I attached the geogrid.log file.

Winston

 

[winstonwu91]

After outputing the diagnostic variables I found that at one specific i (for me i = 158), all perturbation geopotential (ph) is of NaN value, causing dz8w to be NaN value and crashing the sfclayrev calculation. What possibly have caused it? Thank you.

[EDIT] Going further I found out that at the second Runge-Kutta step of the first time step of rk_calc_prep when I think ph is first used (as an input), ph have become NaN in some cores... I checked with wrfinput_d0* files and met_em files and I did not see any NaN values there. It is very strange. I suspect it is a communication problem that somehow ph_2 does not get through?

 

[winstonwu91]

Solved by changing incompatible scheme.

 

[tslin2]

Solved by changing incompatible scheme.

winstonwu91​

I am curious to know what specific scheme you have changed to fix the issue?​

Thanks​