1. CRTM Activities Contributions From: Paul van Delst, David Groff, Quanhua Liu, Tong Zhu, Ming Chen, Emily Liu, Andrew Collard JCSDA Workshop, May 13-15, 2015

2. Overview Version 2.2.x release Features for version 2.3 OSS implementation Issues with current implementation Features for version 3.0 Transmittance coefficient generation package – Not enough time to talk about it, but putting the slides in regardless. JCSDA Workshop May 13-15, 2015 1

3. Version 2.2.x JCSDA Workshop May 13-15, 2015 2

4. Version 2.2.x Released internally for Q1FY16 GSI implementation Feature set: – Overcast radiances. – Turning on the microwave sea surface reflection correction for non-precipitating clouds. – ATMS snow emissivity model. – FASTEM-6 (rather than fix issues with our FASTEM-5 implementation). – Minor source code changes to address zeus meta-server issues. V2.2.0 released 13 April 2015 V2.2.1 released 21 April 2015 – Rolled back the SRF-based ATMS transmittance coefficients. – Parallel test in GSI showed degraded results with new ATMS coefficients. JCSDA Workshop May 13-15, 2015 3

5. Version 2.3 JCSDA Workshop May 13-15, 2015 4

6. Version 2.3 This is the next planned public release. – v2.2.0 was released without some planned features due to time constraints. Feature set: – Cloud fraction – netCDF coefficient datafiles – Update of transmittance coefficient data objects – Generic aerosol optical property object – RT solver computation speedup (see also “Issues” slide) JCSDA Workshop May 13-15, 2015 5

7. Version 2.3 Cloud Fraction Initial implementation (ongoing) – A simple scalar “effective” cloud fraction, C – User is responsible for determining the value – CRTM result ⇒ RTOA = (1-C)*RCLR + C*RCLD – Code is waiting for review – Repeated clear sky calcs ~5% of computational cost Planned implementation – Cloud fraction profile input – The effective cloud fraction will be computed – Multiple overlap schemes (default max-random) JCSDA Workshop May 13-15, 2015 6

8. Version 2.3 netCDF Coefficient Files Switching all coefficient files to netCDF4 – Allows the combination of datasets. – Byte-sex independent. – Easier maintenance. JCSDA Workshop May 13-15, 2015 7 SensorCoeff SensorInfo SpcCoeff TauCoeff ACCoeff (optional) ACCoeff (optional) NLTECoeff (optional) NLTECoeff (optional) SensorInfo group in all files and groups for checking. SpcCoeff Coeff Data SpcCoeff Coeff Data SensorInfo TauCoeff SensorInfo ODPSCoeff ZeemanCoeff (optional) ZeemanCoeff (optional) ODASCoeff ODPSCoeff Coeff Data ODPSCoeff Coeff Data SensorInfo ODASCoeff Coeff Data ODASCoeff Coeff Data SensorInfo ZeemanCoeff Coeff Data ZeemanCoeff Coeff Data SensorInfo Completed CloudCoeff, AerosolCoeff, and various EmisCoeffs also completed

9. Version 2.3 Update of transmittance coefficient data objects There are three reason for doing this: – Removal of pointers from the ODAS and ODPS object definitions. Allocatables provide the same functionality with much less code and zero danger of memory leaks. – It will enable the CRTM to be built in single-precision mode with a simple switch in the build process. This capability was originally designed into the CRTM, but floating point dependencies crept back in via the object definitions. (Review!) – Get the transmittance coefficient objects and their methods in the same design as all the other CRTM objects to allow usage of OO features. The conversion to netCDF4 is also an opportunity to do this. It’s taking a longer than anticipated. – Due to differences between the actual ODAS definition, and a repeated one embedded within ODPS used for water vapour line absorption. JCSDA Workshop May 13-15, 2015 8

10. Version 2.3 Generic Aerosol Optical Property Object Currently the CRTM can only use GOCART-based aerosol optical property definitions. As part of the update of data objects for netCDF coefficient I/O, the AerosolCoeff object is being changed to allow its use with optical property data from any aerosol model. Plan (in concert with the various JCSDA aerosol folks): – Define the data object and file format, used by the CRTM, that will work with any aerosol model output. [easy] – Generate an application that takes aerosol model output (e.g. OPAC) and creates an AerosolCoeff datafile for use with the CRTM. [harder] JCSDA Workshop May 13-15, 2015 9

11. Version 2.3 Cloudy RT Computational Speedup The default scattering radiative transfer solver is the ADA. – It’s fast, but not fast enough compared to simpler algorithms used for “low-stream- count” RT. Slow cloudy computations are impacting applications. – NCEP operational post-processing that generates cloudy radiance product. – MiRS retrieval package. Two requirements: 1.Optimise memory usage of current ADA. Modify internal data objects to not use MAX_N_ANGLES in allocation for more efficient memory usage. This has been started, but on backburner. 2.Addition of dedicated 2- and 4-stream only RT modules. This speed issue really needs to be addressed. JCSDA Workshop May 13-15, 2015 10

12. OSS Implementation JCSDA Workshop May 13-15, 2015 11

13. OSS implementation AER delivered an Optimal Spectral Sampling (OSS) model, which was developed using CRTM v2.0.5. – CRTM-OSS can directly simulate un-apodized radiance – CRTM-OSS computation efficiency depends on applications: fast for smaller ratio of nodes per channel, slower for larger ratio of nodes per channel. Recently, AER delivered new OSS coefficients for AIRS, IASI, CrIS and CrIS high resolution sensors. We have performed a preliminary evaluation of the OSS model in terms of bias and standard deviation. We are working on merging the delivered CRTM-OSS code into the current CRTM trunk. JCSDA Workshop May 13-15, 2015 12

14. OSS Implementation CRTM-OSS unapodised radiance simulation Unapodised radiances provide more spectral information JCSDA Workshop May 13-15, 2015 13

15. OSS Implementation OSS/ODPS speed comparison Clear-sky forward simulation for two gases (water vapour and ozone). L == localised training; G == global training JCSDA Workshop May 13-15, 2015 14 SensorsOSS (seconds) ODPS (seconds) CPU time Ratio IASI_B1_metop-a_L 2260 ch., 1012 nodes 2.7125.7302.11 IASI_8461_metop-a_L 8461 ch., 3680 nodes 8.66222.0062.54 CrIS_npp_G 1146 ch. (B1+B2), 638 nodes 1.6863.3031.96 CrIS_npp_L 1305 ch., 7665 nodes 17.6433.7190.21

16. OSS Implementation IASI B1 OSS/ODPS radiance simulations JCSDA Workshop May 13-15, 2015 15 Profile 1 Warm Atmosphere T skin = 302.56K Profile 2 Moderate Atmosphere T skin = 272.99K Profile 3 Cold Atmosphere T skin = 236.07K Pretty good agreement Large CO 2 and O 3 differences

17. Issues with current implementation JCSDA Workshop May 13-15, 2015 16

18. Issues with current implementation CIRA problem with 3.9μm solar reflectivity when clouds are present Cold temperature bias over winter pole in GDAS Radiative transfer computation speed in scattering atmosphere (mentioned previously, planned for v2.3) – Memory issue – Algorithm implementation JCSDA Workshop May 13-15, 2015 17

19. Issues CIRA-identified SWIR cloudy simulation problem CIRA researchers identified serious problem with 3.9μm solar reflectivity when clouds are present. – CRTM is producing much too cold temperatures, by up to 50K! – CIRA OO (uses SHDOM) produces good results compared to observations. Two changes made – Fix to ADA solar reflectivity application to include solar affected channels. Previously only visible channels had solar reflectivity included. – Updated the infrared ice cloud optical properties (Ping Yang’s group, TAMU) The changes sort of improved results, but not really. The implementation of solar reflectivity in the presence of clouds in the CRTM needs to be fixed. JCSDA Workshop May 13-15, 2015 18

20. Issues CIRA-identified SWIR cloudy simulation problem JCSDA Workshop May 13-15, 2015 19 Data provided by Louis Grasso, CIRA

21. Issues CIRA-identified SWIR cloudy simulation problem JCSDA Workshop May 13-15, 2015 20 Data provided by Louis Grasso, CIRA

22. Issues Cold temperature bias over winter pole This is a potential issue that needs investigation. – Jim Jung (CIMSS/JCSDA) is trying to increase use of water vapour channel information – Added water vapour channels for CrIS and both IASIs – Parallel experiments show a cold temperature bias over the winter pole – Need to assess the performance of the CRTM for infrared channels in cold dry atmospheres to determine if any contribution. JCSDA Workshop May 13-15, 2015 21 Skin Temperature, August 2014, EXP - CNTRL

23. Version 3.0 JCSDA Workshop May 13-15, 2015 22

24. Version 3.0 Nascent feature set – Only mentioning because main interface will change CSEM/CRTM integration – For CSEM science, see Ming Chen’s talk. Change to main CRTM interface – Multiple surface subtypes. – Shift to more complete OO design/usage. An interface description document is being developed. – CRTM requirements. FWD, TL, and AD models. Microwave, infrared, and visible spectral coverage. – CSEM requirements. – Proposed interface to CSEM procedures. JCSDA Workshop May 13-15, 2015 23

25. Version 3.0 CSEM integration Goal is to allow CSEM to be developed completely separately from CRTM. (Duh!) Main issue is getting information into CSEM from CRTM. – Definition of input CSEM data structures need to be made available to CRTM. Two approaches are being considered – Use the CSEM structures as a component of CRTM structures. – Use the CSEM structures as the base and extend them with CRTM structures. JCSDA Workshop May 13-15, 2015 24

26. Version 3.0 CSEM integration CSEM objects as components in CRTM surface structure. TYPE :: CRTM_Surface_type...other components... TYPE(CSEM_Land_Surface_type), ALLOCATABLE :: Land(:)...other components... END TYPE CRTM_Surface_type – Downside here is that the CRTM will be completely dependent upon the CSEM definition. – Adding extra CRTM-specific land info will be difficult. Or square-peg-in-round-hole- y. – This tightly couples the models (A Bad Thing™) JCSDA Workshop May 13-15, 2015 25

27. Version 3.0 CSEM integration CSEM objects as base types for equivalent CRTM objects TYPE, EXTENDS(CSEM_Land_Surface_type) :: CRTM_Land_Surface_type...any CRTM specific components... END TYPE CRTM_Land_Surface_type TYPE :: CRTM_Surface_type...other components... TYPE(CRTM_Land_Surface_type), ALLOCATABLE :: Land(:)...other components... END TYPE CRTM_Surface_type – CRTM definition can change as required – More loosely coupled models JCSDA Workshop May 13-15, 2015 26

28. Closing remarks Huge amount of work performed by David Groff and Emily Liu over the year with CRTM user support, and implementation and testing of CRTM in GSI. – Yong Chen, Andrew Collard, and John Derber too. Releases – V2.3 is planned for Q1FY16. – CRTM-OSS release likely for Q2FY16. Testing schedule in GSI needs to be firmed up. – V3.0 is planned for 2016. Actual date depends on what else is added. Issues mentioned need to be addressed in the meantime. – Same for TauCoeff generation package. JCSDA Workshop May 13-15, 2015 27

29. Transmittance coefficient generation Extra! JCSDA Workshop May 13-15, 2015 28

30. Transmittance coefficient generation package (Taken from Q4FY13 plans!) Process consists of four main steps 1.Reformatting and processing of user-supplied SRFs 2.Generation of transmittance spectra database for profile training set 3.Generation of instrument resolution transmittance spectra (convolution or FFT). 4.Regression of transmittance profiles against the training set to produce the transmittance model coefficients used by the CRTM (TauCoeff’s) Currently this is our single point of failure. Each step needs to be automated and packaged up. Package will allow users to generate their own TauCoeff data. – Suitable for research – Also for organisations who do not wish to distribute their instrument information. JCSDA Workshop May 13-15, 2015 29

31. Coefficient Generation Package Step 1. SRF Processing Process user- supplied SRF data oSRF SpcCoeff User SRF SensorInfo This portion of the process is currently prototyped in IDL. – This allows us to easily reformat data and view results, but not suitable for distribution. – SRF processing code will be translated to Fortran2003. The current oSRF data format (netCDF3) will be converted to netCDF4 – More efficient storage of multiple channel data. JCSDA Workshop May 13-15, 2015 30

32. Coefficient Generation Package Step 2. Transmittances Conversion to LBLRTM (IR) or MonoRTM (MW) input files LBL input Training profiles Spectral Database LBL calculation Profiles Absorbers Angles LBL calculation Profiles Absorbers Angles Transmittance data reformat LBL processing script and utilities TauSpectra This process really just needs to be automated better and packaged. – There’s a fair amount of “housekeeping” software involved. – Also need to take into account batch processing software differences. Users will be responsible for obtaining and installing the third-party LBL code and database. – But we will have to provide the scripts that “glue” them together to generate the data correctly. TauSpectra stored in netCDF4. Currently, we are still using Rosenkranz LBL model for the microwave. Need to switch to MonoRTM. An LBL utility callable from our own processing code (CLBL?) will greatly simplify the LBL processing. JCSDA Workshop May 13-15, 2015 31

33. Coefficient Generation Package Step 3. Convolution/FFT Convolution (broadband), or FFT (FTS) Convolution (broadband), or FFT (FTS) TauSpectra SensorInfo oSRF TauProfile This process is the simplest. Just requires update of all the housekeeping software to handle the new data structures. The instrument resolution transmittance profiles, TauProfile, will be converted from netCDF3 to netCDF4 format. JCSDA Workshop May 13-15, 2015 32

34. Coefficient Generation Package Step 4. Regression Regression of transmittances against profiles TauProfile SensorInfo SpcCoeff This process is well defined. Main issue is linear algebra library, and replication of results with different compilers, different platforms, etc. This is where we can also test impact of single precision on transmittance model algorithm. Training profiles TauCoeff FitStats JCSDA Workshop May 13-15, 2015 33