GETKF calculation in two steps

[metdyn]

Description

This PR sets up a two-step calculation to compute analysis from GETKF with linear observer. The GETKF code can run on 4 Milan nodes with 32 ensemble members at C90 resolution when 30 obs files are each thinned by 75%. Aircraft_temperature and wind obs are commented out, because there are unknown jedi source code issues that goes out of the scope this PR.

• Variational bc is used and reading JEDI source code shows why this works in two steps calculations.

Dependencies

• Vertical localization parameters in ETKF (L2)  #729

Impact

[metdyn]

I am still working on node assignment for executable runs from this PR.

[metdyn]

Temporarily withhold this code change to add more improvements.

[metdyn]

I have been inspecting the input and output from the ensemble runs when this PR was initially created and closed. Now I donot see problems with this infrastructure complement. Hence I will have a few other PRs related to local ensemble DA intended to merge into this base PR.

[metdyn]

Hi all,

I want people to review this PR. It is technical and affect local ensemble solver only. I want to get this PR and a few other GETKF related PR merged to develop, so that we can test the scientific part of the algorithm related to JEDI code. Currently @rtodling and me are having problems with the output from local ensemble solver, and I am not concerned with the swell implementation here.

@mranst mentioned the T2 test failed and I think this is related to the ongoing UKMO work on UFO code

• https://github.com/JCSDA-internal/ufo/pull/4015
• https://github.com/JCSDA-internal/ufo/pull/3772
  which I wrote in - Localization function in the correlated classes throw an exception #705

I modified the yaml input for GETKF to include only sondes, sfcship, atms_n20 and this should make the GETKF tier1 and tier2 test run without error.

[mranst]

I have a limited understanding of the technical aspects of this PR, but from my perspective it looks good. Tier 1 tests ran without issue.

[mo-owenaljabar]

@mranst mentioned the T2 test failed and I think this is related to the ongoing UKMO work on UFO code

@metdyn @mranst what's the configuration that's giving the failure? The amount of time for localisation to be implemented depends on which correlated R class you're trying to use, as we're treating the different classes one by one. Work on cross variable covariances is currently underway, but within group covariances and the new diffusion based R matrix are further off.

[mranst]

@mranst mentioned the T2 test failed and I think this is related to the ongoing UKMO work on UFO code

@metdyn @mranst what's the configuration that's giving the failure? The amount of time for localisation to be implemented depends on which correlated R class you're trying to use, as we're treating the different classes one by one. Work on cross variable covariances is currently underway, but within group covariances and the new diffusion based R matrix are further off.

Tier tests are not currently failing due to localensembleda. I tested the suite against today's build of JEDI and it executed correctly, as far as I could tell.