Yannick opened the meeting announcing that today we would hear presentations on two models, CRTM and RTTOV, from three speakers, Zhiquan (Jake) Liu, Ben Johnson and David Rundle.
Jake started by presenting the following slides on his comparative study of CRTM and RTTOV. Before proceeding to the presentations, he also reminded participants that a code spring is currently taking place on adding tests. So, be aware that there have been many pull requests merged this week. If you see test failures in bundles, you may try wiping your build directory and doing a clean build.
Following is a summary of the presentation (please see slides for details).
Jake presented a study he did for the PANDA-C (Prediction AND Assimilation for Cloud) project which is a joint effort between NCAR, JCSDA and the Air Force with the aim of improving cloud analysis and forecasting. The work presented was conducted in April and May of 2020, and included the development of a C++ interface for RTTOV which allowed the simulation of cloudy radiances for ABI, AHI, AMSU-A and MHS sensors within MPAS-JEDI. This interface enabled the comparison between RTTOV and CRTM.
Jake show results (comparisons between RTTOV and CRTM) for NOAA18 MHS and AHI (see slides for plots and details). In general RTTOV and CRTM showed similar results, more so with microwave than infrared. Differences in the results were attributed to differences in the treatment of hydrometeor ice species by RTTOV and CRTM, and to differences in the handling of surface emissivity.
Jake also experimented with two NWP microphysics schemes (WSM6 and Thompson schemes) that demonstrated notable differences in the production of ice hydrometeors. The message here is to take caution with when using different microphysics schemes.
One more result was presented (final slide) which demonstrated the importance of providing good initial conditions along with including cloudy sky simulation.
Dan asked about the runtime performance of CRTM versus RTTOV. Jake responded that both are fast and there is not much difference between the two (as long as the "fast solver" option is selected in RTTOV).
Ben asked if CRTM was used to simulate the visible band. Jake responded "no" and that he wanted to make the interface general before conducting further tests. Ben also asked if the RTTOV C++ interface was checked into GitHub, and Jake said "not yet" since he wants to clean up the code before publishing it.
Ben noted that CRTM has had good improvement since the April/May timeframe of this comparison study.
Ming-jeong Kim asked how the particle sizes were determined. Jake responded that the effective radius representing the size distribution in the NWP microphysics scheme from MPAS and MPAS-jedi is passed to the radiative transfer models for table lookup. Ming-jeong commented that the differences seen when using different microphysics schemes are likely due to different size distribution representations. She also noted that the lookup tables in CRTM are also sensitive to ice density which may be another source of differences seen in the comparison with ice species.
Greg asked if RTTOV and CRTM are handling a given hydrometeor size the same, and Jake responded that it is likely the two models are handling hydrometeor size differently.
Next up was Ben Johnson who presented the following slides on CRTM development
Here is a summary of Ben's talk (please see slides for details).
Ben started with some background about CRTM. It has been around since 2004, has had lots of contributors and current development includes collaboration with the RTTOV developers. CRTM is currently private but will become open-source, open access and available on GitHub soon. A public release is scheduled for October 2020 (Version 2.4.0). And a spring 2021 release is also in the works (Version 3.0.0).
CRTM is a 1D column model, that handles aerosol, cloud and precipitation scattering (unpolarized) and absorption, along with gaseous absorption and emission. Version 2.4.0 is introducing parallelization using OpenMP (multi-threading), more sensors (IASI, Severi, GOES-17 ABI, eg.), scattering table updates and aerosol updates. Version 3.0.0 will be introducing a polarization solver and ultraviolet support (IQUV), improvements for cloudy radiance, improved surface emissivity and improvements related to SW/IR and aerosols.
Ben presented a few case studies (see plots, details in slides) that demonstrate the motivation for improving cloudy radiance and adding in the polarization capability.
Ben finished up with descriptions of some of the methods and algorithms inside CRTM. Some notes from this section are that the CRTM developers determined that doing the OpenMP parallelization over columns was more effective than doing the parallelization over channels. A 3 parameter Gamma distribution (version 3.0.0) for hydrometeors works well for taking different ice hydrometeor shapes into account. Ben also showed some information about aerosol properties and results from satellite radar/lidar simulation that is under development.
Jake suggested that the ~12 K differences shown on the slide advocating the scattering polarization capability were likely due to surface influence instead of particle scattering. Ben responded that since the data on this plot were sampled for high rain rate only, the surface influence would be insignificant, and the results therefore show the importance of including particle scattering polarization.
Jake asked if there were plans for polarization in RTTOV and Chawn Harlow responded "yes" but polarization won't be available for a while (RTTOV version 14).
Emily asked if the CRTM 2.4.0 features were fully validated yet, the answer was "no" and Emily then asked if her group could help with the validation. Ben of course responded "yes" and indicated that the beta version of 2.4.0 will be available next week. Ben will notify Emily when the beta version is available.
David Rundle (UKMO) then presented the following slides on RTTOV development.
Here is a summary of David's presentation (please see slides for details).
RTTOV provides a Fortran interface and is currently in use at the Met Office. RTTOV is a 1D transfer model like CRTM. The primary focus for RTTOV is to support the UKMO operations, and in the long term the developers would like to provide interoperability with CRTM and have RTTOV included in a "toolbox" of radiative transfer tools.
A pull request to UFO will be coming soon that introduces RTTOV using its Fortran interface. This will include support for microwave instruments and clear sky simulation for both non-linear and TLAD model interfaces. RTTOV version 13 will be released soon (October, 2020), containing the aforementioned capabilities, and as mentioned before the cloudy sky feature will be in the following release, version 14.
The next steps for RTTOV development include testing and replicating SatRad output which are in continuous development. In the October timeframe, plans are to introduce hyper-spectral sensor support (infrared with grey cloud support) and scattering for microwave, infrared and visible bands. In the December timeframe, the plan is to introduce microwave and infrared emissivity atlas support.
Jake asked if the unification of the RTTOV and RTTOV_SCATT systems would be in version 13, and the response was that this will be introduced later in the version 14 release.
Yannick noted that the RTTOV developers need to decide whether to maintain both Fortran and C++ interfaces. This is not urgent but is important to be addressed at some point. David responded that both interfaces are in use now, but they may be able to converge on one interface later on.