Generate a regional static BE by interpolating BJ's global BE

[guoqing-noaa]

RRFSx/rrfs-workflow_beta#76

[TingLei-NOAA]

@guoqing-noaa So, this will be treated , at least at the beginning, as an interpolation (in horizontal and maybe, vertical) process from the global domain to regional domain? As I learned from last mpas tutorial workshop, MPAS develop team has been developing similar tools but not released. Staring from JEDI 's existing functions/interface is maybe a pathway forward to be considered.

[guoqing-noaa]

@TingLei-NOAA BJ is helping us with this. He sent us some codes to use.
The horizontal interpolation is ready in his code. But more effort will be needed to do vertical interpolation.
The conus12km case uses the same vertical levels as BJ's global BE, so we don't need vertical interpolation for now.

[TingLei-NOAA]

@guoqing-noaa That is great!. Hope this tool can be used as a starting point to convert RRFS's current GSI BE variances to mpas's format.

[guoqing-noaa]

@TingLei-NOAA I think Masaroni is working on porting GSIBE to JEDI, right?
We don't work on GSIBE. BJ's global BE is generated from global MPAS forecasts.

[TingLei-NOAA]

@guoqing-noaa Masroni is working on porting regional GSI BE function to JEDI , of course, Using the GSIBE of jedi, jedi could read GSI BE "parameters" rather straightforwardly .
Theorectially, GSI BE "parameters" are related to GSI in their format, but their values (variance and length scales ) are related to the forecast models. So it has been discussed (including Sijie, Ming , Shun and myself and more) to find/develop the way/tool to convert current GSI BE parameters to be able to be used by JEDI (bump, for being now).
Sorry for the confusion in my previous comments.

[Junjun-NOAA]

I have successfully interpolated the three statistics (cov_rh, cov_rv and stddev) from global domain to CONUS 12KM domain and local NICAS files are also generated. However, we had problem with generating the VBAL files. We are now able to use global VBAL files to run experiments. Here is the results from single obs test.

Currently, to use the global VBAL files, we can only use 256 MPI tasks. We are working to figure out how to generate regional VBAL files and the number of MPI tasks is not necessarily 256.

[Junjun-NOAA]

Another issue is that the standard deviation is too small (for example, the max T standard deviation is about 0.48) and hence we got tiny increments when doing pure 3DVAR.

[TingLei-NOAA]

@Junjun-NOAA Thanks for this update. For small values of error , we might try manual tuning.
For VBL , we might start from no balance part in the B for our regionial runs.

[guoqing-noaa]

Thank @Junjun-NOAA for the updates! We will need the 120 mpi layout so that we can use it in the workflow/RDASApp.

@TingLei-NOAA We will use the global VBL first until we get a reasonable regional VBL.
Yes, for the standard deviation, we will inflate that.

[Junjun-NOAA]

An update to this issue. Now I can generated VBAL files using 120 MPIs. And I have tried different inflation factors for the standard deviation. Using a factor of 3 and NICAS, VBAL files generated for the regional domain, the single-obs is performed. Here is the temperature increment at level 20, a preliminary result. More turnings are needed to investigate. Thanks.

[TingLei-NOAA]

@Junjun-NOAA That is great for you to have the balance in bump to be run in your case. I hope I could understand more on how it works soon with your test case a good starting point.
How to use the balance is also dependent on what analysis variables are to be used in our regional mpas-jedi.
Stream function and velocity potential are still not available for regional domains according BJ.
@ShunLiu-NOAA @hu5970 and @guoqing-noaa I think an issue could be opened for the selection of analysis variables for mpas-jedi now.

[Junjun-NOAA]

Since the inflation factor inflates the standard derivation of each variable with the same value, we did not use it but adjusted the standard derivation of each variable separately. For this same case, we have made several adjustments to the standard derivation and correlation length to produce more reasonable results. Below is the temperature increment obtained by inflating the standard deviation of T by 3, stream function and velocity potential by 1/2, as well as reducing the correlation length to 1/4 of the original values that are generated from the global SBE.

[Junjun-NOAA]

After upgrading mpas-jedi module to v3.0.1 and using the current mpassout files in RDASApp, we run the 3dEnVar, 3DVAR and Hybrid experiments using the same SBE in previous comments, we got the following:

[Junjun-NOAA]

In previous comment, the U/V wind is not smooth. A new parameter smooth interpolation factor is introduced in saber to make NICAS give smoother fields, see JCSDA/saber@fd81071 or https://github.com/JCSDA-internal/saber/pull/937. This functionality is already included in the recent RDASApp submodule update PR #217.

The default value for the smooth interpolation factor is 0.25. After trying different values, it is found that the larger the smooth interpolation factor, the smoother and smaller the increment, but longer the effective horizontal lengths, see below:

Choosing the factor of 0.5, the analysis increments for 3DVAR, 3DEnVar and Hybrid runs are shown below: