Error - max column in satinfo file for IR

[rtodling]

I am puzzled about the error-max column in the sat-info file in GSI's recent versions; this is the 6th column in the file. Is it correct to set the max value to a value that is smaller than the ops-error itself (4th column)?

The corresponding JEDI yaml files have the same settings - consistently so (which is good) - but I wonder if GSI and JEDI do different things w/ these numbers (in case it is for some odd reason meaningful to see the error-max to a value larger than the error itself).

Can someone please appease my mind in that this is not an error - or a possible issue in:

a) GSI itself
b) JEDI doing things different than GSI

?

Thank you.

[gmao-wgu]

This column is used to provide the upper bound in the gross-check, that is, if abs(omf) > minium(3*obs error, err_max), the observation will be tossed away.

[gmao-wgu]

Yes, JEDI is consistent with GSI regarding err_max, except for the q observations from the rawinsonde. There is no gross check in the YAML, which I believe Akira knows more about.

[rtodling]

Yes, JEDI is consistent with GSI regarding err_max, except for the q observations from the rawinsonde. There is no gross check in the YAML, which I believe Akira knows more about.

pls, let's leave the handling of conventional obs for another discussion - I have seen the issue in swell and that's fine.

[rtodling]

Yes, JEDI is consistent with GSI regarding err_max, except for the q observations from the rawinsonde. There is no gross check in the YAML, which I believe Akira knows more about.

This column is used to provide the upper bound in the gross-check, that is, if abs(omf) > minium(3*obs error, err_max), the observation will be tossed away.

right - which means that for all cases where err-max has changed recently, the criterium above is always going to be:

abs(omf) > err_max

and because err_max in the latest changes is typically smaller than 3*obs_err we are effectively letting more data into the analysis which have larger abs(omf) but still giving those data the same obs-err ... is this really what we want to do? It seems to me that if we are letting data with more scatter in (larger omf) we should down weight the data no?

[gmao-wgu]

I am not sure what you mean. My understanding was that we only have a small number of high-peaking channels from IR(AIRS, IASI and CrIS-FSR) with decreased err_max values, which results in fewer observations being assimilated.

[rtodling]

BTW: I just want to say that it looks to me like - for IR - the error bound check done in JEDI is not as simple as what you state above:

   out[ichan][iloc] = std::fmin(3.0 * obserr[ichan] 
                           * (1.0 / pow(errflat[0][iloc], 2))
                           * (1.0 / pow(errftaotop[ichan][iloc], 2)), obserr_bound_max[ichan]);

in routine:

filters/obsfunctions/ObsErrorBoundIR.cc

of UFO ... perhaps someone understand clearly the relationship between the JED and GSI tests ...

[gmao-wgu]

That is because the final upper bound used in the gross-check is not a constant value provided in the satinfo file, and it also depends on other things.