search

NOAA-EMC / RDASApp

Regional DAS
GNU Lesser General Public License v2.1
1 stars 7 forks source link

potential options to create a GSI baseline test mechanism for MPASJEDI #99

Open guoqing-noaa opened 6 days ago

guoqing-noaa commented 6 days ago

GSI cannot read MPAS model output directly, so it is not as easy to compare MPASJEDI vs GSI as we can for FV3JEDI vs GSI.

There are two potential routes to establish some comparison between MPASJEDI and GSI:

  1. Use MPASSIT or a similar process to interpolate MPAS model output to regular lat, lon grid to connect GSI @Junjun-NOAA has successfully run MPASSIT and converted MPAS to a CONUS12km HRRR-like WRF output. But MPASSIT does not do a vertical interpolation and regional GSI can only work on sigma (or hybrid sigma) vertical coordinate. @spanNOAA tried to do vertical interpolation and successfully converted the HRRR-like WRF output to 53 HRRR vertical levels. However, GSI has been stuck at some point and this point jumps for different reruns. We will need more time to resolve it if we decide to go on this route.

  2. Use RAP/HRRR to cold start an MPAS case and do a comparison between MPASJEDI and GSI Currently, init_atmosphere_model does not output pressure fields in init.nc. We will need to update the source code to add this capability.

ShunLiu-NOAA commented 6 days ago

@guoqing-noaa Even after converting MPAS grid to the GSI analysis grid or the HRRR-like grid, we are unable to compare the analysis results between MPASJEDI and GSI. I believe that the best way to go is to identify flaws with FV3JEDI and then update MPASJEDI with the results. After we've gained enough confidence, we can begin cycling DA and compare the analysis results or forecast to RRFSv1 or HRRR.

guoqing-noaa commented 6 days ago

@guoqing-noaa Even after converting MPAS grid to the GSI analysis grid or the HRRR-like grid, we are unable to compare the analysis results between MPASJEDI and GSI. I believe that the best way to go is to identify flaws with FV3JEDI and then update MPASJEDI with the results. After we've gained enough confidence, we can begin cycling DA and compare the analysis results or forecast to RRFSv1 or HRRR.

@ShunLiu-NOAA Thanks for the discussion. I agree that FV3JEDI will help in some situations. On the other hand, regardless of JEDI's "model-agnostic" claims, MPASJEDI is different from FV3JEDI in many ways. We still need to establish some mechanism to compare MPASJEDI vs GSI.