bartbrashers
New member
I'm trying to test the effect of using soil moisture nudging in Australia using the PX LSM and ACM2 PBL scheme. I have two runs that are identical expect for the values of pxlsm_smois_init and pxlsm_soil_nudge:
(That's the beginning of the with-nudging 5-day run, subsequent runs will re-start from it and use pxlsm_smois_init = 0)
For the with-nudging, I get the common RIBX never exceeds RIC, RIB(i,kte) error after 22 simulated hours, almost halfway between the 4th and 5th 6-hourly metoa_em files, in d02 (a least that's the last line in rsl.error.0000). d04 would be the next loop, but d04 is only 100x100 points, and the RIBX happens at 57,159, so it could be in d02 or d03 which are both 160x160. It's suspect that it's about halfway between metoa_em files, which I'm using for analysis nudging.
For the no-nudging case, I can run the the 5.5-day inits simultaneously because each does not depend on the previous. Out of 15 5.5-day overlapping runs, 12 or so had that error (or CFL errors in d01 over Tasmania). With-nudging runs used 64 cores, no-nudging runs each used 8 cores, with the same WRF-4.2 executable.
Previous posts say to look for bad inputs. Using ncview to look at the metoa_em files, I don't see anything particularly out of the ordinary. The location of the RIBX error is usually in d02 (DX=9km) near I,J = 57,159, about 75km (d02) or 12km (d03) from the coast, with no surface OBS nearby (so probably not an SST problem, or a bad OBS problem). That is, however, a cell at the very top boundary of my domain - but d01 near there looks fine to me as well.
I did notice that the field SOILTEMP looks blocky near the coast, but smoother inland, and the point I,J = 57,159 is near the edge of one of those blocks. ST0000007 does not look blocky, nor does SKINTEMP. I don't see SOILTEMP in METGRID.TBL, so I'm not even sure where it comes from or what it gets used for, nor how to make it less blocky.
I did try using two different SST datasets (FNMOC and RTG) but both tests used ERA5 as inputs. Both had the same RIBX problem.
Any hints on what I should be looking for? Does the NaN for w-star imply too much convection?
Code:
# grep -e bl_pbl -e sf_sfclay -e sf_surface SM_nudge/2017-10-18/namelist.input.wrf
sf_surface_physics = 7, 7, 7, 7, 7,
sf_sfclay_physics = 7, 7, 7, 7, 7,
bl_pbl_physics = 7, 7, 7, 7, 7,
# diff_namelist SM_nudge/2017-10-18/namelist.input SM_no_nudge/2017-10-18/namelist.input
< pxlsm_smois_init = 1, 1, 1, 1
> pxlsm_smois_init = 0, 0, 0, 0
---
< pxlsm_soil_nudge = 1, 1, 1, 1
> pxlsm_soil_nudge = 0, 0, 0, 0
(That's the beginning of the with-nudging 5-day run, subsequent runs will re-start from it and use pxlsm_smois_init = 0)
For the with-nudging, I get the common RIBX never exceeds RIC, RIB(i,kte) error after 22 simulated hours, almost halfway between the 4th and 5th 6-hourly metoa_em files, in d02 (a least that's the last line in rsl.error.0000). d04 would be the next loop, but d04 is only 100x100 points, and the RIBX happens at 57,159, so it could be in d02 or d03 which are both 160x160. It's suspect that it's about halfway between metoa_em files, which I'm using for analysis nudging.
Code:
# tail -5 rsl.error.0000
Timing for main (dt= 11.86): time 2017-10-18_09:50:55 on domain 4: 0.16249 elapsed seconds
Timing for main (dt= 11.86): time 2017-10-18_09:51:07 on domain 4: 0.15057 elapsed seconds
Timing for main (dt= 11.86): time 2017-10-18_09:51:18 on domain 4: 0.15237 elapsed seconds
Timing for main (dt= 35.59): time 2017-10-18_09:51:18 on domain 3: 0.98769 elapsed seconds
Timing for main (dt= 71.17): time 2017-10-18_09:51:18 on domain 2: 2.58357 elapsed seconds
# tail -5 rsl.error.0058
d04 2017-10-18_06:00:00 Input data processed for aux input 9 for domain 4
-------------- FATAL CALLED ---------------
FATAL CALLED FROM FILE: <stdin> LINE: 425
RIBX never exceeds RIC, RIB(i,kte) = NaN THETAV(i,1) = 197.1508 MOL= -18.71513 TCONV = NaN WST = NaN KMIX = 4 UST = 0.5062621 TST = -0.6880610 U,V = -3.220325 5.207067 I,J= 57 159
-------------------------------------------
For the no-nudging case, I can run the the 5.5-day inits simultaneously because each does not depend on the previous. Out of 15 5.5-day overlapping runs, 12 or so had that error (or CFL errors in d01 over Tasmania). With-nudging runs used 64 cores, no-nudging runs each used 8 cores, with the same WRF-4.2 executable.
Previous posts say to look for bad inputs. Using ncview to look at the metoa_em files, I don't see anything particularly out of the ordinary. The location of the RIBX error is usually in d02 (DX=9km) near I,J = 57,159, about 75km (d02) or 12km (d03) from the coast, with no surface OBS nearby (so probably not an SST problem, or a bad OBS problem). That is, however, a cell at the very top boundary of my domain - but d01 near there looks fine to me as well.
I did notice that the field SOILTEMP looks blocky near the coast, but smoother inland, and the point I,J = 57,159 is near the edge of one of those blocks. ST0000007 does not look blocky, nor does SKINTEMP. I don't see SOILTEMP in METGRID.TBL, so I'm not even sure where it comes from or what it gets used for, nor how to make it less blocky.
I did try using two different SST datasets (FNMOC and RTG) but both tests used ERA5 as inputs. Both had the same RIBX problem.
Any hints on what I should be looking for? Does the NaN for w-star imply too much convection?