Seemingly random spurious GRIB output (v3.8)

[GlennCreighton]

Hi WRF gurus!

It's been a long time since I've asked you a question, but recently we've had reports of spurious data in our gust diagnostic from one of our WRF (v3.8) domains. The issue would happen randomly, about once per cycle. We tracked the issue down to random levels of VGRD that were being output with spurious values of ~25 m/s higher than it should everywhere.

rec 239:24479756:date 2018111506 VGRD kpds5=34 kpds6=107 kpds7=9815 levels=(38,87) grid=255 sigma=0.9815 27hr fcst:
VGRD=v wind [m/s]
timerange 0 P1 27 P2 0 TimeU 1 nx 252 ny 279 GDS grid 3 num_in_ave 0 missing 0
center 57 subcenter 1 process 11 Table 133 scan: WE:SN winds(grid)
Lambert Conf: Lat1 27.930000 Lon1 59.390000 Lov 52.550000
Latin1 60.000000 Latin2 30.000000 LatSP -90.000000 LonSP 0.000000
North Pole (252 x 279) Dx 5.000000 Dy 5.000000 scan 64 mode 136
min/max data -24.2 7.3 num bits 9 BDS_Ref -242 DecScale 1 BinScale 0

rec 240:24558980:date 2018111506 VGRD kpds5=34 kpds6=107 kpds7=9735 levels=(38,7) grid=255 sigma=0.9735 27hr fcst:
VGRD=v wind [m/s]
timerange 0 P1 27 P2 0 TimeU 1 nx 252 ny 279 GDS grid 3 num_in_ave 0 missing 0
center 57 subcenter 1 process 11 Table 133 scan: WE:SN winds(grid)
Lambert Conf: Lat1 27.930000 Lon1 59.390000 Lov 52.550000
Latin1 60.000000 Latin2 30.000000 LatSP -90.000000 LonSP 0.000000
North Pole (252 x 279) Dx 5.000000 Dy 5.000000 scan 64 mode 136
min/max data 1 32.7 num bits 9 BDS_Ref 10 DecScale 1 BinScale 0

rec 241:24638204:date 2018111506 VGRD kpds5=34 kpds6=107 kpds7=9645 levels=(37,173) grid=255 sigma=0.9645 27hr fcst:
VGRD=v wind [m/s]
timerange 0 P1 27 P2 0 TimeU 1 nx 252 ny 279 GDS grid 3 num_in_ave 0 missing 0
center 57 subcenter 1 process 11 Table 133 scan: WE:SN winds(grid)
Lambert Conf: Lat1 27.930000 Lon1 59.390000 Lov 52.550000
Latin1 60.000000 Latin2 30.000000 LatSP -90.000000 LonSP 0.000000
North Pole (252 x 279) Dx 5.000000 Dy 5.000000 scan 64 mode 136
min/max data -25.3 7 num bits 9 BDS_Ref -253 DecScale 1 BinScale 0



This happened two cycles later, this time at the first sigma level (note: the above was output from GPOST, which destaggers the VGRD output, but the below is the staggered data from the model itself, which still shows the problem so this is not just a GPOST issue):

rec 124:4857219:date 2018111518 VGRD kpds5=34 kpds6=119 kpds7=9985 levels=(39,1) grid=0 0.99850 (ETA level) 1140min fcst:
VGRD=v wind [m/s]
timerange 10 P1 4 P2 116 TimeU 0 nx 252 ny 280 GDS grid 3 num_in_ave 0 missing 0
center 57 subcenter 1 process 220 Table 133 scan: WE:SN winds(grid)
Lambert Conf: Lat1 27.907000 Lon1 59.388000 Lov 52.550000
Latin1 60.000000 Latin2 30.000000 LatSP -90.000000 LonSP 0.000000
North Pole (252 x 280) Dx 5.000000 Dy 5.000000 scan 64 mode 136
min/max data 0 37.9 num bits 9 BDS_Ref 0 DecScale 1 BinScale 0

rec 125:4936718:date 2018111518 VGRD kpds5=34 kpds6=119 kpds7=9945 levels=(38,217) grid=0 0.99450 (ETA level) 1140min fcst:
VGRD=v wind [m/s]
timerange 10 P1 4 P2 116 TimeU 0 nx 252 ny 280 GDS grid 3 num_in_ave 0 missing 0
center 57 subcenter 1 process 220 Table 133 scan: WE:SN winds(grid)
Lambert Conf: Lat1 27.907000 Lon1 59.388000 Lov 52.550000
Latin1 60.000000 Latin2 30.000000 LatSP -90.000000 LonSP 0.000000
North Pole (252 x 280) Dx 5.000000 Dy 5.000000 scan 64 mode 136
min/max data -28.8 11.5 num bits 9 BDS_Ref -288 DecScale 1 BinScale 0

rec 126:5016217:date 2018111518 VGRD kpds5=34 kpds6=119 kpds7=9885 levels=(38,157) grid=0 0.98850 (ETA level) 1140min fcst:
VGRD=v wind [m/s]
timerange 10 P1 4 P2 116 TimeU 0 nx 252 ny 280 GDS grid 3 num_in_ave 0 missing 0
center 57 subcenter 1 process 220 Table 133 scan: WE:SN winds(grid)
Lambert Conf: Lat1 27.907000 Lon1 59.388000 Lov 52.550000
Latin1 60.000000 Latin2 30.000000 LatSP -90.000000 LonSP 0.000000
North Pole (252 x 280) Dx 5.000000 Dy 5.000000 scan 64 mode 136
min/max data -29.7 10.8 num bits 9 BDS_Ref -297 DecScale 1 BinScale 0

It seems to me that perhaps the BDS_Ref (scaled minimum value) is randomly being set improperly…the data looks correct, just with an improper reference minimum, as though some byte were added/removed somewhere (in fact, each time this happens the BDS seems to be off by ~ 256, which is a suspiciously magical number).

Have you guys ever run into this before?

Thanks,
Glenn

 

[kwerner]

Hi Glenn,
I am aware that since posting this, you have discussed the issue with a member of our WRF support team. As this is an open forum, I wanted to post the specifics of the advice given to you, in case it's helpful to anyone in the future:

As no one in our group uses the grib output option for WRF, we are not experts on the matter, but see a couple of different strategies you could use:
1. If your team is at all interested in pursuing this issue directly (changing source code), let us know and we can make sure that you have the proper techniques with git and github, so that your efforts can be folded back into the standard code.
2. If you are interested in perhaps addressing this in a round about fashion (output select fields in a different format, and then translate that netcdf to grid later), we can chat about how to do that also.

It may be beneficial to determine whether this is the only field that exhibits this behavior. If no other fields are impacted, does stuffing the same data into an existing and unused array yield the same symptomology?

If possible, let us know what is determined. Thanks!