use of the surface pressure rather than delp in real world fv3jedi var applications

[TingLei-NOAA]

Hi, Rdasapp fv3jedi folks,

In the current rrfs fv3jedi ctests in the rdas app, the delp rather than surface pressure is used as an analysis variable.
It is ok when our ctests are supposed to demonstrate the functions are verified.
( I set that for that is easier to create the required bump setup)
However, in our real work applications, as in our targeted workflow, we should start from surface pressure, following GSI variational app does in rrfsv1, rather than this aggressive use of delp.
I think this practice works for both fv3jedi var and fv3jedi enkf.
This issue would be supposed to facilitate discussions on this topic.

[SamuelDegelia-NOAA]

@TingLei-NOAA Thank you for raising this point. In most of our fv3-jedi yamls, except for the GETKF, we set both air_pressure_thickness and air_pressure_at_surface as analysis variables. Do you know if it will cause a problem to list both delp and surface pressure?

Note that the 3DEnVar yaml and ctest was only recently updated to include air_pressure_at_surface last week. So if you are using a slightly older version of RDASApp then you might not have this update.

[TingLei-NOAA]

@SamuelDegelia-NOAA seems

RDASApp/rrfs-test/validated_yamls/templates/basic_config/fv3jedi_en3dvar.yaml

Line 6 in aa801a0

analysis variables: &3dvars [eastward_wind,northward_wind,air_temperature,air_pressure_thickness,water_vapor_mixing_ratio_wrt_moist_air,cloud_liquid_ice,cloud_liquid_water,ozone_mass_mixing_ratio,air_pressure_at_surface]

is using delp as the analysis variable.

Did you mean to use surface pressure as the state variable? Use of surface pressure as the state variable while using delp as the analysis variables won't change much regarding the aggressiveness using delp as the analysis variable.

Using both delp and surface pressure in the state variables should produce identical results using only one of them as the state variables ( conceptually) which might deserve some comparisons to explore the inner conversions between those pressure related variables. Maybe another issue could be opened for it.
But, to use both as analysis variables would be definitely suspicious and unless special treatments are implemented and prove its "rightness".

[SamuelDegelia-NOAA]

@TingLei-NOAA I might be missing something but I see both air_pressure_thickness and air_pressure_at_surface defined in line 6 there as analysis variables. Does air_pressure_thickness (delp) overwrite air_pressure_at_surface?

[TingLei-NOAA]

@SamuelDegelia-NOAA My apologies! I just missed that :).
Conceptually, I don't think surface pressure and delp should be used together as the analysis variables. Regarding fv3jedi coding, I also suspect the current fv3jedi are not ready to treat this scenario though it might not report an error.
I would strongly suggest to take this opportunity to remove delp from the analysis variables and let surface pressure as the pressure analysis variables.
I would add the context : In fv3 and fv3-jedi, surface pressure is calculated from the summation of delp. In GSI and fv3-jedi, when surface pressure is used the analysis /control variable, the delp would be finally updated according an hydrostatic relationship from the surface pressure increment.

[SamuelDegelia-NOAA]

@TingLei-NOAA I agree that using both air_pressure_thickness and air_pressure_at_surface could cause problems. Do you plan to do any comparisons between them in fv3-jedi, or do we want to update the yamls now?

[TingLei-NOAA]

@SamuelDegelia-NOAA I think unless we have some plans (for some reasons) to implement the function using delp and ps together as analysis variable, I don't feel it deserve to compare that option. But maybe, comparison between using delp or ps is interesting but it is hard to get convincing results to show that delp rather than the "traditional" ps should be used.
I would recommend to fix the simultaneous use of delp and ps as the analysis variable by removing delp to avoid any future use in real cycling experiments.

[HuiLiu-NOAA]

@Masanori-NOAA : what do you think about this?

[Masanori-NOAA]

In GSIbec, ps is used as the analysis variable. There should be no problem with removing delp from the analysis variables for 3DVar.