Virtual temperature flag gives problematic values in read_prepbufr.f90

[jak5808]

While verifying self-cycled ENSDA experiment results, I found that the virtual temperature flag vtcd in read_prepbufr.f90 gives a nonsense value (either nan or 3x10^-38). This error leads to large positive biases in temperature observation values, as temp. obs that should be flagged as virtual are instead designated sensible. It also leads to issues with humidity readings.

I was able to rectify this bug by declaring vtcd to be an integer as opposed as a float in subroutine read_prepbufr. It appears the NOAA-EMC/GSI acknowledged this problem here: NOAA-EMC/GSI#642 . In their case, they simply updated CMakeLists.txt, specifically from bufr::bufr_4 to bufr::bufr_d . One of these two solutions should be implemented immediately, as this bug has a substantial impact on the analysis.

[XuLu-NOAA]

Thanks for reporting this issue to us!

[XuLu-NOAA]

@jak5808 I have added a fix to this integer issue. Would you mind helping review the changes? 8094411

[jak5808]

Hi Xu,

I made these changes to my code, built successfully and ran the analysis successfully. vtcd now gives a reasonable value, and the resulting O-B biases as reported in output files are much smaller, well within what is to be expected.

[XuLu-NOAA]

Hi Xu,

I made these changes to my code, built successfully and ran the analysis successfully. vtcd now gives a reasonable value, and the resulting O-B biases as reported in output files are much smaller, well within what is to be expected.

That's great to hear! Thanks again for reporting this and testing for the fix!