2020-08-24

Agenda Item 1:

AOD discussion

Discussion:

 Yingtao: Hongli, how are you going to test CMAQ AOD?

Hongli: We will have to see.

Yingtao: CMAQ in CRTM is not exactly CMAQ. You will have to map it from 64 to 8 species. Do you have any experience with the mapping?

Hongli: No, I will probably need help.  I will have to have a look at the aerosol types in CMAQ 12. Ho can help with the mapping.

Yingtao: I have version 4.0 with 29 species. I got some email communication between Mark Liu and Sarah maybe 10 years ago when they developed version 2.0 of CMAQ and I need to revisit those emails. The other question I have is, do you assimilate AOD or radiance?

Hongli: AOD.

Yingtao: So, the control variable is microphysical variables?

Hongli: I think it’s concentration.

Hongli: What do you mean directly assimilate AOD?

Yingtao: When you set up the CRTM you need to set up the solar angle. AOD is much simpler.

Hongli: I saw some other papers with AOD that have a lot simpler forward operators. We want to use the CRTM to assimilate AOD. I can check out your code to see how 29 aerosol species work. In the meantime, the best person to contact is Mark Liu.

Yingtao: Once I give you the code I can set up the forward calculation.

Hongli: Also, this month I have another project and can have a look at the code but …

Yingtao: I’m also calculating coefficients for ABI.

Ben: Hongli, are you a UCAR employee?

Hongli: I was at UCAR for several years, but now I’m at NOAA.

Ben: Do you know Rajesh Kumar? He’s the one we should talk to. And Mariusz.

Hongli: I talked to Mariusz and he is one of our collaborators. I will also talk to Kumar for AOD assimilation.

Ben: When NASA does AOD assimilation they also don’t use the CRTM.

Hongli: Yes, for the CRTM you also need to consider a lot of technical issues.

Yingtao: The scattering table is to convert the aerosol properties

Ben: You want to use the CRTM for a number of points: you want to know how the AOD is impacted by the microphysics. Second, our final goal is to assimilate radiance for AOD. So what Cheng is doing is making the CRTM more useful for this purpose.

Hongli: I am looking forward to work with all of you.

Results:

CRMT CMAQ implementation is finished.

Tasks:

-       CMAQ -> CRTM CMAQ mapping needs to be written.

-       CRTM AOD test case needs to be written.

Responsible People:

Hongli, Yingtao

Deadline:

October 2020 (REL-2.4)

Agenda Item 2:

CHYM Discussion

Discussion:

 Yingtao: Ben, I saw you have been working on the cloud scattering table for a while.

Ben: For a long time. Are you talking about the CHYM? We wrote a paper on it. It is supposed to go into REL-3.0. Cheng is working on the NetCDF interface. We want to use something similar for the cloud tables. I am working closely with Greg Thompson. He is developing the UFO interface and he is also working on the microphysics. So, we will use this to propagate microphysics into the CRTM. The next step is trying to use GFDL microphysics. There’s no guarantee that it will make things better. But in the long run we want to be able to have a set of scattering properties based on the microphysics that gives you the best O minus B.

To answer your question: we haven’t made much progress in the past few months but we will do more work in the time before REL-3.0. Does this answer your question?

Yingtao: So, you already have all the properties for all the habits?

Ben: We have tables but they don’t have spectral consistency between MW and IR. We want to use the same set of particle habits from MW to UV, so you don’t get a bias by switching spectral bands.

Yingtao: Very ambitious. So, you want to have a single table that works for the whole spectrum. The current one is separate, right?

Ben: Yes, and it’s difficult to test. The other thing is microphysics, when you are able to match the properties in the model and the CRTM. We are pretty well set with the scattering properties

Ben: Yingtao, the particles from Guosheng Liu were they delivered to you?

Yingtao: I think they were delivered to Mark Liu.

Ben: We want to get away from the Liu database because the results are biased, for instance to get the right mass the snowflake size had to be 10cm in some cases. What we will do is just get a mass-dimension relationship and get the scattering properties that match this relationship. As soon as Patrick wraps up the coefficient generation we will probably have a code sprint and you are welcome to join.

Yingtao: I think that takes a long time.

Ben: We have long term and intermediate term goals.

Results:

Basic cloud scattering framework exists.

Tasks:

Scattering properties need to be integrated into the CHYM framework.

Responsible People:

Ben, Patrick, Greg Thompson

Deadline:

Release of CRTM REL-3.0

Agenda Item 3:

REL-2.4 Discussion

Discussion:

 Yingtao: For REL-2.4 I have a suggestion. We should better make a major feature its own release number.

Ben: So, you mean something like OpenMP 2.4 and NetCDF 2.5?

Yingtao: Yes.

Ben: I agree, but at the same time I have already permission to do the REL-2.4 as it is.

Yingtao: What about particle sizes?

Ben: One thing is to extend particle sizes out to ten microns. The other part for particle scattering is just to be consistent with the GFDL scattering properties complete with backscattering coefficients. It seems that Mark Liu implemented my backscattering code.

Ping Yang provided some tools and I also have a good relationship

Kevin: I just wanted to continue the discussion about the REL-2.4. My impression was that the MOB wasn’t quite ready for the release and there are some concerns on the STAR side which we can work through. We haven’t really seen much in terms of results with the new scattering tables. The second thing is that we haven’t identified any testing opportunities, especially all-sky assimilation with MIRS. The main purpose for us was to get the AOD part for Hongli out. If all the testing can get out until October that is fine but we can also extend the schedule. We should come to an agreement on a testing schedule.

Ben: So there was no mention of AOD in REL-2.4.?

Kevin: So it’s the CMAQ.

Isaac: Is there any documentation on the cloud scattering properties?

Patrick: Yes, I can send you the papers by myself and Bingqi Yi.

Isaac: Those are the properties in the CRTM?

Patrick: Yes.

Ben: Kevin is it necessary to have a fully vetted REL-2.4?

Kevin: Probably. I don’t know how to define full vetting. But it is necessary to involve the MIRS team in the process and check that there are no unexpected challenges. And instrument teams also have the choice of what to include.

Ben: I can guarantee that there will be changes from bugfixes. You think that is going to be an issue?

Kevin: No, just if it changes the answers.

Ben: The major changes will be the interfaces for aerosols and clouds and then the OpenMP components. There were some modifications that were required by OpenMP.

Kevin: Just to make sure that it doesn’t crash. That doesn’t mean it has to be a CRTM issue.

Ben: Sounds like Yingtao has developed new coefficients for ABI. Will he share that?

Yingtao: Yes. I’m reading your REL-2.4 Github issues. You have a long list. First of all the instruments. Can we do all of this before October?

Ben: Some of this has already been done. I believe Tong had completed some of these. I think almost all of this has been completed.

Kevin: I don’t know how detailed you capture the coefficient generation but we have some that are not in the github. We can put them in there.

Ben: Who do you want to do this, Yingtao?

Kevin: Yes. Sensor by sensor is probably the way to go, even if it is kind of tedious.

Results:

REL-2.4 needs to be tested by STAR before public release.

Tasks:

-       A STAR branch needs to be created on Github.

-       A testing suite for the MIRS team needs to be created.

Responsible People:

Ben, Kevin

Deadline:

October 2020 (REL-2.4)

Agenda Item 4:

Miscellaneous Updates (pyCRTM, GPU, Cheng)

Discussion:

Ben: Any other points to discuss? Bryan are you still involved with the Python development.

Bryan: I haven’t heard anything recently but I can help.

Ben: I haven’t heard from Ed. For me it’s really his responsibility. Daniel, any update on GPUs?

Daniel: Last week I had a discussion with Patrick and Jim. The option of using one GPU thread per profile and that doesn’t seem to fit. So, I am approaching the problem using one OpenMP thread and write a kernel for one of them. I profiled the code and found out that five functions take the most time. Some of them do matrix inversion and I am using a library for that. One problem I have is not to make an error. So a tool I use is PGI PCAST that runs code on the GPU and CPU. And I would like to have a test case. I am approaching the problem by writing a kernel for each OpenMP thread and using PCAST for comparison.

Ben: Can you specify an ideal test case in your Github issue?

Daniel: Yes, sure.

Ben: Cheng, do you have any updates?

Cheng: Previously Mark Liu sent me subroutines for computing coefficients in the CRTM but the challenge is to connect the output of these routines to the CRTM LUT’s. Perhaps the experts can have a look at the output.

Ben: Anybody else? No? So, I have to come up with a sensitive work plan in terms of reasonable effort in the next few months.

Kevin: Ok, I’m probably looking more into the logistical side and the supposed vetting we have to do.

Ben: So, for EMC we created a branch off our Github repository for them. I coordinated with the NCEPlibs group. EMC always likes to add things to the CRTM.

Kevin: Yes, that is probably a good way to do it.

Results:

Tasks:

Responsible People:

Bryan, Daniel, Cheng

Deadline: