Examine the assimilation of radiosonde observations - GSI vs JEDI for the FV3 case

[guoqing-noaa]

I am working on examining the assimilation of radiosonde observations - GSI vs JEDI. The general plan is to do a single observation test using the FV3 test case included in RDASApp. Then I will do a full test of radiosonde DA. After that, I may try the 13km case @spanNOAA has been working on.

[guoqing-noaa]

increments from FV3-JEDI pure 3DEnVAR (2022052619z)

[guoqing-noaa]

increments from FV3-JEDI hybrid 3DEnVAR (2022052619z)

[delippi]

Hi @guoqing-noaa, can you share where you are running your tests? And do you have the corresponding plots from GSI? I just finished validating hofx for mesonet observations (#80) yesterday and here are some key takeaways:

• The GSI and JEDI create geovals differently--I suspect that the difference is a result of methods of horizontal interpolation (e.g., 3 vs 4 points). Therefore, you should use GSI-geovals to feed into the vertical interpolation operator. I've collected and modified some codes to do this. I/we may need to create a branch of these codes (GSI and ufo) for everyone to use more easily.
• Not only are the horizontal interpolations slightly different, there is only one vader recipe for calculating the layer midpoint pressure which is used in the vertical interpolation operator as the vertical coordinate. That recipe is src/vader/recipes/AirPressure_A.cc which uses the Phillips method. For fv3_regional in GSI, I found that it just uses pressure interface average. Just note that they are different and will give different hofxs! I hacked recipe A to use layer average instead, but its probably better to create a whole new recipe and submit a PR in ufo.
• Check out issue #54 for some details.

It might take some effort to get all this set up. Until we get something more permanent in the codes we can just use some of these hacks.

[guoqing-noaa]

@delippi Thanks for the information. I will check the hofx validation later.

[guoqing-noaa]

Single sounding OB test increments from the 2023061012z CONUS_13km FV3MPAS case

Thank @spanNOAA for running the experiments and making the plots!

[guoqing-noaa]

For reference, there are no major differences in DA results between NICAS resolution 8 and 16

[ShunLiu-NOAA]

@guoqing-noaa JEDI h(x) vs. GSI h(x) scatter plot will also help further address the difference between the operators.

[guoqing-noaa]

@guoqing-noaa JEDI h(x) vs. GSI h(x) scatter plot will also help further address the difference between the operators.

Thanks, @ShunLiu-NOAA! We have been focusing on a single observation test now. We will check the hofx validation in the next step when we assimilate all sounding obs.

If we compare the above figures between the RDASAPP fv3jedi case and the 2023061012z test case, we can find that 30 ensembles provide much more flow-dependent information. @TingLei-NOAA