Radiance Bias correction in ODS files

From GEOS-5
(Redirected from Radiance Bias correction)
Jump to navigation Jump to search

Pawson, Steven (GSFC-6101) wrote:

Hi Meta,

Do you know where we keep the bias correction values for the radiance assimilations? I'm thinking of MERRA, I just don't know if these are in the ODS files, or somewhere else.

Also, since I have started, I will ask something more technical. If I understand correctly, O-Fs do not have any bias correction applied, but the O-As do. Is this correct?

Just point me somewhere if this is all documented thoroughly. It's the MERRA system I am interested in for now.

Thanks, Steven

The diag 'bin' files from the GSI record the (O-F) without bias correction and the (O-F) with bias correction. When we write the ODS files we calculate a bias correction value for the radiances via [(omf-without-bias-corr) minus (omf-with-bias-corr)] and store that value in the "xm" slot. The (O-F) recorded in the ODS file is the bias corrected (O-F) value. The obs value recorded in the ODS file is also bias corrected by the ODS conversion software, by subtracting the bias correction value from the original observation.

The ODS processing reads the before-analysis diag file and the after-analysis diag file and processes both in exactly the same way to create two ODS files. These files are merged into one final file where it looks for the obs in the after analysis file that corresponds to the before-analysis obs file. If a match is found, the value of "(O-F)" in the after-analysis file - which is actually the bias corrected (O-A) - is copied into the (O-A) slot of the before-analysis array, and the merged values are written to a new file.

What the above means technically speaking is that all the obs attributes in the ODS file are from the before-analysis diag file - including the bias correction obtained as described above. The only value from the after-analysis diag file is the corresponding (O-A), and that value would be bias corrected using the bias correction state at the end of the analysis which may well not be the same as the value recorded in the 'xm' slot in the ODS file.

I don't know if this is documented thoroughly someplace else - I just took a look at the code to check exactly what was being done and wrote it down here.