On January 23rd the JCSDA team met with our partners at NASA, NOAA, US Navy, US Air Force, and the UK Met Office to review the last three months of accomplishments and discuss goals for the upcoming quarter, and year. Major Q3 accomplishments include completed ionosonde and TEC decoders for space weather, standardization of NetCDF in CRTM, and the addition of sea ice to regular CI.

Observations Team

CRTM v3.1 is being used for all relevant instruments
Continue demonstrating Skylab w/ FV3 and MPAS
Ingest and storage highlights:

For space weather: completed ionosonde, and ground- and space-based TEC (Transmittance and Emissivity Calculation) decoders

For TEC, the old format was TENET: the Navy has helped with the decoder for this file type

Added converter for OMPS (Ozone Mapping and Profiler Suite) Nadir Profiler and Total Column H5 products available through NOAA CLASS (and ingested by partners)

Investigating WeatherBench for verification and monitoring, currently using METPlus
Small updates to TROPICS converter including tomorrow.io products
Ocean surface wind converter updated for Muon, which is now available
NOAA EMC and GSL

Changes to IODA to enable parallel processing with BUFR
“Made a lot of progress in a short time”
Mapping and configuration files are now stored in the EMC SPOC (Specific Preprocessing of Observations Configuration) repository with the new EMC ObsForge repository as the overarching repository for observation pre-processing

“Thanks to partners, a lot of work being done and everyone has been helping each other”
NBAM observation error inflation bug fix
Different cloud treatment in CRTM based on surface type for all sky radiances. Needed to reproduce GSI. (NOAA)
Unified surface correction type being specified by YAML
Missing hydrometeor increment bug fix in FV3-JEDI
From Met Office: Superobbing that does not require H(x), ObsFunction to calculate cloud top pressure using the Stable Layers method, ObsFunction to calculate tropopause height level from model fields
Assessing quality of GNSS-RO Ocean Surface Wind data
USAF commercial weather program is supporting efforts to explore the impact of Windborne Global Sounding Balloons: A sampling campaign was initiated in the western Pacific, with the balloons executing multiple profiles per day. JCSDA is exploring the impact using the JEDI-MPAS interface with ERA-5 as verifying analysis
We’ve been interested in smallsat data for a while, and are now working on the tomorrow.io small microwave sounders data

developed and tested the variational bias correction, and are working to augment the bias predictors

UCAR JCSDA and COSMIC teams have been developing and testing a new GNSS-RO observation error model: This hybrid error model combines dynamic information from the observation using the local spectral width (LSW) in the lower levels (<10km), and a statistical approach for the upper levels (>30km) and traditional treatment between 10-30km.
Bug fix put in for GNSS-RO obs error inflation, further investigation ongoing

Inkind collaborations:

Radar reflectivity at OU/CAPS: Fixed several bugs in the reflectivity operator within UFO, including corrections to air density calculations and in the reading of MPAS hydrometeors in the JEDI-MPAS interface, Conducted a DA cycle assimilating radar reflectivity using EnVar in JEDI-MPAS for a storm case. Next quarter: PR for bug fixes and interface changes to support the reflectivity operator, Implementing power transform in JEDI-MPAS
NASA GMAO: fix for an issue w/fv3-jedi linear variable change that allows passing of 3d-pressure to the observation operators that require it, still requires an adjustment to the generic increment writer to be completed
NRL: Assimilation of column of NESDIS OMPS Nadir Profile and Total Ozone Content products being implemented within the FALCON workflow
NOAA EMC: Acceptance work continues in the context of “end-to-end” processing using the EMC GDASApp, aircraft bias correction capability tested and validated, working on extension of ATMS/AMSU-A processing to include observations with precipitation
NOAA GSL and EMC: LETKF/LGETKF are tested for both surface and upper air observations, 3D hybrid EnVar with regional static Background Error Covariance interpolated from global one, adding regional version of GSI recursive filters as well as Multi Grid Beta Filter (MGBF) to SABER, full set of RRFS observations are converted from BUFR to IODA for further testing, S. Pan (NOAA) has worked to develop correction scheme for surface observations in JEDI
Met Office: JEDI-based processing of radar Doppler wind and reflectivity run in regional model trial,eEach observation type has been updated ready for Met Office implementation of JEDI-BasedObservation Processing Application (JOPA) at the end of 2025, surface regional model processing underway, legacy Met Office QC flags replaced with Diagnostic Flags

Space Weather:

Using PyIRI and working to reproduce the behavior of SAMI3, the non- releasable Navy model
Gauss Markov model, which will be used for forecasting, almost ready
JEDI–PyIRI interface is working well
Have the 1st Hofx profile now, which is at an acceptable level to proceed to the next cycle
Created utilities to generate ensembles and GeoVals interpolators for JEDI-PyIRI ctests and forward operator tests
Generated GeoVals for UFO interfaces to provide interpolated background files at observational locations
Developed initial ctests and test cases to compute H(x) using the generic vertical interpolation operator for Ionosonde electron density observations

Goals for next quarter:

Continue validating, testing, and implementing forward operators and filtering functions to enhance UFO versatility and adaptability
UFO-CRTM development/improvement for spaceborne radar and small satellite microwave
Skylab demonstrator for benchmarking, higher order regression testing and to perform impact studies
Observation impact assessment and interaction with partners for examination of emerging observations
Continue JEDI development for space weather and move on to GNSS TEC operator development and LETKF testing

CRTM

Mostly working towards the release of CRTM 3.2.0
Streamlining integration with JEDI and UFO
Making it easier to run CRTM in operational style
NetCDF is now standard except for some non-user-facing files
Ready to start testing the new release at the end of January
CRTM 3.2.0 features:

Enhance SpcCoeff_BIN2NC to Support Generation of Associated ACCoeff and NLTECoeff NetCDF File
Improve the CMake build system by ensuring paths remain functional for all users, increasing build system flexibility
CRTM Coefficient Tarball Download Logic
All binary coefficients converted to NetCDF format
New optional feature for computing downwelling radiance at user-defined pressure level
New module to load MWwater_Coeff directly from the existing MWwaterCoeff generation packages
Users no longer need to load both cloud and aerosol LUTs for CRTMv3 calculations
SURFEM fast ocean surface emissivity model implementation for MW/IR
Generic Optical Properties Interface
New Sensors: GEOXO-Sounder (GXS, via STAR), tomorrow.io updates (tmos), Orbital Microsystems (gems-2), PolSIR, AMPR, 3dS-INSAT imgr/sndr (via STAR)
the updated ADA solver has been reviewed and merged into CRTM code base, ready to release with CRTM v3.2

PolSIR project:

Instrument hasn’t launched yet, so this prep will let us use the data as soon as it’s released; will be especially useful for cloudy conditions, particularly thin clouds like cirrus
Simulated H(x) has been created using the PolSIR coefficients

General support is currently taking 25-30% of CRTM team time
Coefficient generation:

Coefficient generation really needs 2 more FTEs
Updated all TROPICS coefficients to reflect correct polarization basis, polarization angles, version numbers
Added COSMIR / COSMIR-H development support
Created/delivered PolSIR coefficients
Created/delivered AMPR coefficients
Created tomorrow.io sounder coefficients, which are being validated
Work still needed: Ozone contribution(s) at submillimeter frequencies, visible coefficients for reflectance calculations, proper treatment of polarization basis across scans for various instruments, updates to CO2 profiles used in training, improved constituent training / profiles
CRTM coefficient generation at STAR:

Collaborated with SSEC/Univ. of Wisconsin (Zhenglong Li) and GMAO/NASA (Erica McGrath-Spangler and Min-Jeong Kim) on creating and testing the GXS-GeoXO and IRS-MTG coefficients. Coefficients have been delivered. Work is in progress to evaluate the impact of the hyperspectral instruments.
Reported the coefficient generation work for GXS and IRS on the GXS proxy data working group meeting
Reported the CRTM work for the active sensor and all-sky radiance simulation on the Commercial Data Program (CDP) End of Year (EOY) Review meeting

Early work by STAR on GXS and IRS simulation in CRTM is looking very good
STAR AI work is showing good response in many categories with AI CRTM models

Two AI_CRTM models are designed to meet requirements of CRTM, NOAA retrieval systems, and data assimilation processes: Ocean model and user-defined-emissivity (UDE) model
AI_CRTM BT predictions are more than 20 times faster than CRTM, AI Jacobian calculations for both DNN and ResNET models are faster than CRTM
The functionality of AI_CRTM for BT and Jacobian predictions will be extended to include IR and hyperspectral sensors in the future

ML Emulator CRTM (Lucas Howard)

Looking promising so far
Fully integrated and resembling CRTM with similar inputs and outputs–the plan is to incorporate the emulator to assist with the computations where applicable

Future tasks:

Standardize and improve the NetCDF interface for all coefficient LUTs (version #, release #, generic I/O modules)
Offline packages in support of CRTM development and tests
UV/Visible/IR reflectance models for snow/ice
Aerosol coefficient generation package
Initial pivot toward AI front-end development for CRTM (CRTM v4.x). Coordinated effort with JCSDA partners, CU-Boulder

SOCA

AOP24 priorities:

Prepping for operational use at NASA GMAO and NOAA EMC
Scientific advancement–4DEnVar, coupled ocean/atmosphere DA

Hybrid LETKF-3DEnVAR

Fixed bugs to allow use of full 80 ensemble members for NOAA
Removed the last legacy SOCA static B component and moved into SABER block
Next quarter will work on continued CPU / memory / IO profiling and optimization

Advancements in sea ice DA

Getting rid of JCSDA-specific repos and now using authoritative upstream repos, which will be very helpful for future maintenance
Sea-ice now exercised in CI
Ice pack utilities now do “sanity check” for sea ice
Torch library now available in gnu version, so you can turn on and test AI ice balance; next quarter will be investigating for scientific validity

HTLM with SOCA

Kriti Bhargava has started working on running HTLM with SOCA in a ¼ degree regional test case
So far working as expected, and will be doing cycling experiments next quarter

Contributions from NOAA EMC:

Hybrid EnVAR added to global workflow
sea-ice analysis improvement
Next quarter: Preparation for GFSv17 C1152/0.25 hybrid EnVAR test

Contributions from NASA GMAO:

Steady progress on ocean/sea-ice DA improvements: in testing, JEDI version of this showing much better performance than current operational system
Next quarter: investigating SOCA T/S T/S/SSH balance Jacobians

Coupled DA with OASIM:

Continuing to add FV3 atmospheric variables to the coupled OASIM h(x), including aerosols
In testing and hopefully next quarter will have working coupled Hofx OASIM

Priorities for next quarter:

Operational: Global ¼ deg hybrid-LEKTF/3DEnVAR marine DA and cleanup, profiling, and optimization
Scientific: 4D ocean DA, coupled ocean/atmosphere DA

Communications

LinkedIn is doing well, averaging several hundred views per post
Jedi-Docs update is a key project for next quarter
In-kind affiliation reporting guidelines have been standardized
Presentation and publication processes are on the JEDI github wiki
YouTube is also doing well, all presentations from conferences hosted by JCSDA are posted to the channel
If you find an error in documentation, please open an issue and/or report it

COMPO

Mostly focused on 4DEnVAR for trace gas with TEMPO

Running cycling GEOS-CF/JEDI experiments with dual resolution c360/c90 with various ensemble size (Aug 2023)
Tuning the DA system and fixing bugs
Adding independent observations (ground-based and airborne) to SkyLab for evaluation
Adding documentation
Large increment has been showing in the morning for TEMPO: we looked into it and found it may be due to greater uncertainties with the larger solar zenith angles and input that, which resulted in a more realistic value
Vertical increment showed larger increments close to the upper troposphere, did some tuning in the obs operator to get more realistic increments

Added obs:

AirNow, in-situ measurement of NO2, O3, and PM across CONUS with official site classification
Pandora, column NO2 (and formaldehyde) measurement across the globe
Both now in SkyLab in monitoring mode for systemic validation

Ran SkyLab experiments using data from AEROMMA and STAQS field campaigns

Helpful to compare these with TEMPO measurements, the data will be brought into TEMPO analysis

Infrastructure

Steve Grace joined the team this quarter
In the past R2D2 was all local, which created some issues with data management and updates; the R2D2 API will run on a web server, allowing for better version control and stability
R2D2 API Progress:

Industry standardized HTTP REST API implemented using OpenAPI 3.0 Spec (OAS)
Lightweight Python client wrapper created and integrated with EWOK/Skylab
E2E local testing with low res and toy models completed successfully with Skylab
Next quarter will switch to the new system

Ingest and data quality progress:

Now in the ingest suite: MRMS Reflectivity, Pandora NO2, MODIS AOD, CYGNSS, DPR, airnow
In progress: Continuous database consistency, data store synchronization, and data storage utilization monitoring tools
In progress: User friendly R2D2 functions for managing archive ingest requests

Increasing automation for data quality this AOP

Improving optimized experiment scrubbing across platforms

Spack-stack:

spack-stack-1.9.0 release in progress
Supporting Intel’s LLVM compiler 2024.1 with ifort
JCSDA will not support new Apple OSX releases
1.9.0 will be used to support UFS workflows
In February we will do an internal JEDI stabilization sprint targeted at 1.9
One year lifetime for spack-stack releases; dropping 1.6 support with this release
Collaborative effort, lots of time from lots of people across core team and partners

JEDI Workflow:

Updated to a new JEDI Workflow scheme to offer more flexibility for organizing your directory structures
Automated build scripts in jedi-tools for jedi-bundle and workflow applications
Simplified instructions for building Jedi-Skylab using jedi-tools

Testing

Ongoing maintenance of JEDI CI system and incremental platform work
spack-stack platform testing template using Jenkins

Interfaces and Optimization

Major themes for AOP24:

Improvements to generic components
Optimization of generic code targeting CPU, memory, GPU offload

GPU offloaded B-matrix

Developed a draft GPU offload capability for interpolation in Atlas
Investigated feasibility of using ECTrans GPU offload and identified missing adjoint code
Updated instructions for building Spack-Stack and JEDI with NVHPC toolchain
Next quarter: Finalize GPU interpolation in Atlas with Atlas development team, develop adjoint of GPU inverse-SHT in ECTrans

IODA:

Created simple filtering application, primarily for testing and demo
Working on a more flexible obsspace representation in memory: Steve Vahl took this over from Phil Underwood when he left the Met Office
Completed the initial version of data frame container: More flexible memory representation of obs data, optimizes data access from UFO client
Once fully implemented and tested, this container will replace the current ObsGroup container within ObsSpaces
Next quarter: Populate OSDF container with data from IODA reader

VADER:

Identified use-cases for VADER Linear Variable Changes that can be better supported; a design is being finalized to address these requirements

Linear variable change algorithms is coded, but not the corresponding non-linear algorithm because it isn’t needed
The increment passed to the variable change functions doesn’t include all the same variables as the first guess state passed to the trajectory function
The increment variables are not even a subset of the first guess variables
Avoid calculating and/or storing trajectory fields that will not be needed

Algorithms

Yannick Tremolet is currently acting team lead
Major algo themes AOP24:

Multiscale background error covariances with Skylab-HRDAS (Nate Crossette)
Ensemble of Data Assimilation with Skylab-GEOS (Clémentine Hardy Gas)
Continuous Data Assimilation with Skylab-HRDAS (Christian Sampson)
Generic development and maintenance

Multi-scale localization

Starting testing at high resolution
ProcessPerts added to workflow

Ensemble of DA with Skylab-GEOS

cycling EDA of 32 members with entire observing system from GMAO
Plans to compare with current ensemble in use at GMAO

Continuous DA with SkyLab-HRDAS

Added ability to add new obs between outer loops
Added ability to extend window at each outer loop and add obs that were outside the original window
Added ability to shift beginning of the window in progress; work on this will continue next quarter
Added ability to forecast from analysis in variational application

EWOK: added capabilities in SkyLab:

New io pool based IODA reader can be tested in the workflow
New observations in the ingest suite (see INFRA presentation)
GEOS restarts ingest suite
Multiple outer loops for experiments
Ensemble Kalman Filter suite

Congratulations to the entire JCSDA staff, our partners, and in-kind contributors for the tremendous progress and achievements of these milestones!

Photo by NOAA on Unsplash