2020-06-25

Agenda Item 1:

Status Update by Ben

Discussion:

Introduction by Ben:

A lot of things need to be discussed. REL-3.0.0 alpha is behind. Ben is migrating the CRTM JEDI repo to CRTM_dev; Corresponding CMake modifications are substantial; CMake will allow to add unit tests and regression tests. Ben explains what unit- and regression tests are. Outside developers make it necessary to increase code coverage; Discussions about creating REL-2.4.0 with OpenMP support. OpenMP across channels didn’t work because the Jacobians don’t behave correctly. Ming Chen’s CSEM should also be part of CSEM from Ming Chen; The development of REL-2.3.1 stalled because the subversion repo failed.

If people are interested in spectroscopy, Ben has contact to HITRAN developers at Harvard to discuss future spectroscopic requirements.

Questions:

Hui-Ya Chuang: Will REL-2.4.0 be bitwise identical to previous versions?

Ben: Yes, except for bugs that were found.

Hui-Ya: Will the regression tests include applications, such as GSI?

Ben: No, application developers are responsible for that.

Hui-Ya: We test brightness temperature calculations already and can provide the tests.

Ben: Good.

Sarah Lu: This is system is not part of JEDI tests?

Ben: The goal is that the regression tests will also be called from JEDI ctest, but they are also standalone.

Haixia: What is the target date for REL-2.4.0?

Ben: There’s no target yet because it’s not part of the AOP, but it should arrive soon.

Haixia: Are you going to document the impact of the bugfixes?

Ben: The bugfixes substantially change some outputs, so there will be heavy documentation.

Hui-Ya: Is there a time frame?

Ben: Probably 6 months but this is a preliminary discussion.

Hui-Ya: We will be happy to test the new modifications. It did delay our OpenMPI tests.

Hui-Ya: Do you plan to use MPI?

Ben: GSI is calling CRTM one profile at a time. UPP has additional flexibility. UPP loops over profiles, but that loop already exists in CRTM.

Yingtao: A lot of users do parallelization outside of CRTM. The major difference to your implementation is the overhead from initializing the CRTM. Have you done an impact study?

Ben: The code changes were substantial but not significant, but the timing is a system dependent issue. There’s an issue of memory allocation that might be less for using the internal profile loop. For this reason, it is also important to develop timing tests. Also when you use OpenMP, load-balancing in MPI is much improved. Ben is happy to cooperate for UPP update.

Mariusz: What is the impact of the parallelization on UFO operators with loops over profiles, e.g. AOD?

Ben: I wasn’t aware of AOD looping over profiles outside of CRTM. If the answer is yes this should probably be moved into the CRTM.

Yingtao: Cmake will be included in REL-2.4.0? Each component is a big change. Shouldn’t this be included in 2.5.0, 2.6.0, … and so on?

Ben: This is still a discussion without plans. The idea is to have this in a version of 2.4, but these comments will be considered.

Result:

Feedback on a possible REL-2.4.0 is welcome

Tasks:

-       Finish migrating CRTM JEDI repository

-       Include CMake into CRTM_dev

Responsible People:

Benjamin Johnson

Deadline:

December 2020

Agenda Item 2:

Coefficient Generation Progress (Patrick Stegmann)

Discussion:

Talk:

Coefficient generation is more or less done. Remaining tasks are:

-       Documentation

-       UV/VIS coefficients.

Comments:

Ben: Thanks for all the work.

Yingtao: Will there be an open coefficient generation package?

Patrick: Yes, we plan to publish it on the CRTM_coef repository.

Tom: Are you able to do interferometers?

Patrick: Transmittance in Fourier space is available. Next steps are apodization and inverse Fourier transform.

Ben: We need to look into the different apodization approaches and which one is the most common, e.g. the Hamming function.

Jim Jung: I suggest caution. There are several different methods.

Ben: Working with the instrument team would be the best decision.

Result:

New MW and Thermal IR coefficients can be generated.

Tasks:

-       Publish coefficient generation package and documentation.

-       Look into VIS, NLTE, and Zeeman splitting

Responsible People:

Patrick Stegmann

Deadline:

End of AOP

Agenda Item 3:

Aerosol Updates

Discussion:

Talk:

Working on an updates refractive index database. Major parts will be including the CMAQ aerosols. Then NetCDF I/O should be included. Yingtao has provided the CMAQ data.

Comments:

Sarah Lu: What would be the content of a look up table?

Cheng: I’m using the CMAQ 5.0 version and we are looking to create an aerosol coefficient package.

Mariusz: Are you updating GOCART as well?

Cheng: Yes. I did a comparison between the current CRTM input and the current GOCART version.

Mariusz: We want to use the same or similar table as NASA is using, because FV3-CHEM already uses the same aerosol parameterization as GEOS.

Cheng: In the future we want to accommodate different types of aerosol inputs.

Sarah: The binary luts need to be updated to NetCDF. Do you also provide backscattering coefficients in the lut?

Cheng: Probably.

Yingtao: We have long and short-term plan. We have a table for GOCART and CMAQ for the short term. So there are two separate versions of CRTM. In the future users should only have one version. You can compute the backscattering coefficient from the phase function expansion coefficients.

Sarah: Cheng has the latest GOCART lut from NASA and we hope Cheng can make the CRTM lut consistent.

Yingtao: Using the CRTM aerosol spacing requires some sort of mapping for users. Maybe this can be updated.

Mariusz: Having DA for too many species doesn’t make sense because they will be highly correlated. For this reason, there is no need for too many species in CRTM.

Ben: We will start with the GOCART species, but with NetCDF comes additional flexibility. Our concern is that for UV and Far IR sensors, new species will be active that we haven’t considered. We want flexibility for the future.

Mariusz: I have a general question about GOCART. There are 5 bins for dust. In CRTM you have to assume a distribution within these bins. What distribution did you assume?

Sarah: A lot of the original information comes from Goddard.

Mariusz: In nature there are no bins but a continuous distribution. What is the physical nature of what is done in the CRTM?

Cheng: I can only guess what the distribution is, probably lognormal. Yingtao might be able to answer this.

Yingtao: There’s no documentation.

Result:

The newest version of GOCART has been obtained and compared to the existing CRTM data.

Tasks:

-       Update GOCART

-       Test CRTM NetCDF interface

-       Look into aerosol size distributions

Responsible People:

Cheng Dang

Deadline:

End of AOP