Mpas-jedi be

We are looking to utilize JEDI for Hybrid-EnVAR applications, hence the requirement of a static BE matrix. For that, we assumed that the NMC method could be used for the generation of the static matrix. One of the test configuration files referenced [48-24hr] perturbation netcdf files from what I believe are retrospective 00Z, 06Z, 12Z, 18Z MPAS simulations that were used as input to BUMP. Some of the lines look like:

  • filename: /glade/scratch/bjung/ncl_product/NMC_gfs_f24_f12/2018030100/PTB_f48mf24.nc
  • filename: /glade/scratch/bjung/ncl_product/NMC_gfs_f24_f12/2018030106/PTB_f48mf24.nc

Are the “PTB_f48mf24.nc” files (48H and 24H lead times) from the same forecast? Or is one lead time from an earlier forecast but valid at the same time as the other?

Do the input files to BUMP have to contain the pre-computed perturbations (PTB_f48mf24)?

Do the input files to BUMP have to contain all the fields from the complete MPAS “restart” files? Or can the fields simply be a subset (stream_function, velocity_potential, temperature, spechum or relhum, surface_pressure, qc, qi, qr, qs, qg) and any additional mesh information be contained in a separate netcdf file?

Is it a requirement for BUMP to have continuous periods of input data? Or can it handle missing dates?

Hi @jpcipria, welcome on the JCSDA forum. I can answer some of your questions:

  • I think that for the NMC method, forecasts are valid at the same time.
  • Input files for BUMP can be:
    • An ensemble of states if available (in the yaml section ensemble)
    • An ensemble of pre-computed perturbations (ensemble pert)
    • Two ensembles of states for the NMC method (ensemble base and ensemble pairs)
  • I don’t know what fields are required in MPAS files (you can ask Bjoung-Joo Jung - bjung@ucar.edu).
  • Inputs in BUMP can come from any set of cycles, BUMP does not care about the period continuity.
1 Like

Hi @jpcipria ,

Here I am adding one of the answered questions through email.

Also, thank you @benjaminmenetrier for answering most of @jpcipria 's questions!

> Do the input files to BUMP have to contain all the fields from the complete MPAS “restart” files? Or can the fields simply be a subset (stream_function, velocity_potential, temperature, spechum or relhum, surface_pressure, qc, qi, qr, qs, qg) and any additional mesh information be contained in a separate netcdf file?

We don’t need to use the “restart” format. The sample files for B training can be a subset of the larger file.

There is a separate stream to define these subset variables mpas-jedi/stream_list.atmosphere.control at develop · JCSDA/mpas-jedi · GitHub

Then, in the yaml keys for ensemble read, you need to set the additional key “stream name: control”.

I don’t know if this part is included in the release version of code.

Thank you,
BJ