BUFRReaderBUFRDC.get_value() method - incorrect values

[jintaglee]

Hello,

I have a Bufr file from which I want to pick out the values of a specific descriptor. However, it appears the values look wrong. In order for this problem to be reproduced I have uploaded a sample data file in the following anonymous FTP server,

Name of FTP server: ftp.bom.gov.au Login name: anonymous Password: (not required) Location of Bufr file: /anon/home/cawcr/10day/jinlee/HIMCSR_1.bufr

Following is the set of commands I used to pick the values of the descriptor number, 96 ('BRIGHTNESS TEMPERATURE') from the message=2 and subsets=1, 2, etc.,

bfr=BUFRReader('HIMCSR_1.bufr') bfr.get_next_msg() bfr.get_next_msg() bfr.get_value(96,1)

• bfr.get_value(96,1) returns 1.7e+38 and this is correct.
• bfr.get_value(96,2) returns 291.30000000000001 but the value should be missing
• bfr.get_value(96,3) returns 292.19999999999999 but the value should be missing
• etc.

I used ECMWF's Metview as an independent check on the values. I checked subsets of other messages and there there doesn't appear to be any discrepancy between the values printed out by get_value() and Metview. So the problem seems to be for message=2 and some of its subsets.

If you can have a look I would appreciate it.

Regards,

Jin

[jdkloe]

Hi Jin, thank you for reporting your problem. I succesfully downloaded your sample BUFR file and if I decode it I can reproduce the same output that you get. However, I am not convinced the output is wrong. I also decoded this BUFR message using the latest ecCodes version (v2.0.0). I executed: ./bufr_dump -j f -t /path/to/HIMCSR_1.bufr > tmp and on line 7347 of the generated tmp file I find: "null, 291.3, 292.2, null, null, ... etc" for "brightnessTemperature". Also if I look at a longer range of subsets the values produced by ecCodes and pybufr_ecmwf exactly match, so I am convinced I am looking at the same bufr message with both tools.

How sure are you about the Metview output? Maybe it uses a different indexing? Please note that pybufr_ecmwf starts counting at 0 for elements in a subset, but starts counting at 1 for subsets. Its confusing, I know, and I hope to get this cleaned-up in a next version.

Anyway, it may be easier to get the data using the expression: data = bfr.msg.get_values_as_2d_array()

[jintaglee]

Hi Jos,

Sorry for this late reply. I was tied up with an operational matter and didn't have the chance to have a look at the problem.

You're right: pybufr-ecmwf is decoding the Bufr message correctly. I checked using EMOS library and the values are identical to those from pybufr-ecmwf. So this means Metview is wrong (?!), which I find difficult to believe since it is a software package used by quite a lot of people at ECMWF. I will have to contact ECMWF and find out if it's Metview or it's something I'm doing wrong.

Thank you, Jos. I will keep you informed of what I find out from ECMWF.

[jdkloe]

Hi Jin, thanks for your message. Please keep me posted when you find out what was the issue here. I also don't understand why Metview would give different results, since it is build on the same bufrdc library from ECMWF.

[jdkloe]

For now, closing this issue.