NASA/GMAO Contributions to GSI Ricardo Todling Global Modeling and Assimilation Office GSI Workshop, DTC/NCAR, 28 June 2011 • • • • OUTLINE GSI Infrastructure New Instruments Methodologies Closing Remarks Contributions from: A. da Silva, A. El Akkraoui, W. Gu, J.Guo, D. Herdies, W. McCarty, D. Merkova, M. Sienkiewicz, A. Tangborn, Y. Tremolet, K. Wargan, P. Xu, & B. Zhang Questions/Comments: Ricardo.Todling@nasa.gov Ongoing Development • GSI Infrastructure: – – – – – Revisit ChemGuess_Bundle Introduce MetGuess_Bundle Generalize Jacobian Introduce interfaces to GSI-Jacobian/CRTM for Aerosols and Clouds Revisit interface to TLM and ADM for 4D-Var • New Observation Types and State-Variables: – – – – – MOPITT SSMI CrIS and ATMS OMPS Doppler Wind Lidar • Methodologies: – – – – Use of cloud-cleared moisture background to assimilate IR instruments GMAO-GOCART Aerosols influence on radiance assimilation Add Bi-CG minimization and corresponding Lanczos pre-conditioning Estimation of tendency-based Q (system error covariance) GSI Infrastructure Revisit ChemGuess_Bundle Introduce MetGuess_Bundle Generalize Jacobian Introduce interfaces to GSI-Jacobian/CRTM for Aerosols and Clouds Revisit interface to TLM and ADM for 4D-Var GSI Infrastructure: ChemGuess and MetGuess Bundles • GSI_Chem_Bundle renamed to ChemGuess_Bundle • Introduce MetGuess_Bundle as a means to ingest meteorological guesses into GSI: – presently working for clouds-related fields – being extended to work with basic fields (u, v ,tv, etc) • anavinfo file: – Updates made to chem_guess table – Add met_guess table to control contents for MetGuess_Bundle • Future work includes: – Instantiation of ChemGuess and MetGuess Bundles GSI Infrastructure Interfaces to Aerosols & Clouds • Adding aerosols and clouds to Guess Bundle allows for these to be passed to CRTM; parameter in anavinfo tables determines what’s to feed to CRTM and how. • Add flexible interface to allow for user-specific controls to handle aerosols and clouds (see Tutorial) Interface to AD/TL models • Revisit to support ESMF • Available interfaces exist now for at least three global AD/TL models: – GEOS-5 FV-dynamics – GEOS-5 FV-cubed-dynamics – NCEP Perturbation model New Instruments MOPITT Carbon Monoxide SSMIS CrIS and ATMS OMPS O3 (OSSE-like) Doppler Wind Lidar (OSSE-like) New Instruments: MOPITT CO MOPITT - Measurements Of Pollution In The Troposphere Changes entail: - mild change to obsmod - add usual suspects when handling new observing types, e.g.: - readCO - setupCO - intCO - stpCO - Estimate and set B(co). (from Andrew Tangborn) • Four profiles of MOPITT CO are randomly placed on the globe and assimilated using GSI. Preliminary results are consistent with shape of averaging kernel. • Cycling experiments are on the way. New Instruments: OMPS O3 (OSSE) OMPS – Ozone Mapping and Profiler Suite • High Fidelity Measurements: -Total column (like TOMS) -Vertical profiles (like SBUV) • OSSE Setting: -Generate truth: MLS-O3 & OMI/TC -Simulate Radiances – Forward RT -Apply Instrument Models -Retrieve Profiles -Assimilate Retrievals (GEOS-5 DAS) -1 degree resolution (from Philippe Xu) Results show: - Data are ingested into GSI at all levels - QC control works (but rate of rejection can be adjusted) - Analysis works effectively - Penalties are in good range - Time series show fast convergences - OMA and OMF are all very small and OMA are smaller than OMF New Instruments: OMPS O3 (OSSE) OMPS – Ozone Mapping and Profiler Suite a) 5 hPa b) 100 hPa Analysis error (%) of retrieved ozone assimilation from TRUTH - At 5 hPa errors are small in most of region; orbit tracks of OMPS analysis are noticeable. - At 100 hPa errors are large where retrievals are most difficult: Tropics as the ozone value are very small (<0.1ppmv). (from Philippe Xu) New Instruments: OMPS O3 (OSSE) OMPS – Ozone Mapping and Profiler Suite Retrieved vs MLS TRUTH (%) OMPS sampled vs MLS TRUTH (%) Monthly Zonal Mean analysis errors • The results show that OMPS data agree well with MLS in the stratosphere and in most of the troposphere. • In the tropical UT and LS there is large discrepancy (%) between MLS and OMPS, where the ozone mixing ratio are very small (<0.1 ppmv); needs more work. (from Philippe Xu) New Instruments: Doppler Wind Lidar (OSSE) • Measurements ESA/Aeolus: -Rayleigh backscatter (clear sky) -Mie backscatter (clouds/aerosols) • OSSE Setting: -ECMWF Nature Run (NR) -Errico’s simulated observations -Simulated obs: -KNMI Lidar Perf Anal Simul (LIPAS) -LOS: GEOS-5 replay with GOCART forced with NR -Experiments assimilate -DWL (Rayleigh and Mie) -Rayleigh only -Mie only -1/2 degree resolution (from Will McCarty) Results show: -Diminished impact toward surface - less observations - large contamination - Nearly neutral in NH/SH - winds larger determined by balance New Instruments: Doppler Wind Lidar (OSSE) Changes entail: -mild change to obsmod -And typical - read_lidar - setupdw - intdw - stpdw Reduction in RMS by adding DWL (from Will McCarty) Increase in RMS by adding DWL New Instruments: Doppler Wind Lidar (OSSE) Results indicate: - Upper-troposphere - Mie impact neutral away from tropics; mildly positive in tropics - Rayleigh impact positive throughout; dominates in tropics - Lower-troposphere - Mie and Rayleigh give redundant impact: either provides all information - All-in-all OSSE tends to over-state impact of observing system - Obs error need to be better adjusted (esp. for Mie) (from Will McCarty) Methodologies Use of cloud-cleared moisture background to assimilate IR instruments GOCART Aerosols influence on radiance Bi-CG minimization and Lanczos pre-conditioning Estimation of tendency-based Q (model error) Methodologies: Cloud-cleared q variable for IR Changes entail: - add cloud frac to guess - cloud frac to crtm_interface (water-vapor) Picture displays mean OmF for AIRS calculated using full q variable (red) and cloud-clear q variable; some reduction in bias is observed when new is used – results are still preliminary. (from Dagmar Merkova & A da Silva) Methodologies: Aerosol Radiance Contamination • • CRTM allows for the inclusion of (GOCART) aerosols The GEOS-5 GOCART aerosol species have been introduced as state variables in GSI AOD Validation MISR – No aerosol analysis for now – Aerosol effects included in the observation operators for IR instruments: AIRS, HIRS, IASI, etc • Control Experiment: – – – – Fully interactive GEOS-5 GOCART aerosols Standard global GSI ARCTAS period: Summer 2008 Resolution: ½ degree GEOS-5 • Aerosol Experiment: – Fully interactive GEOS-5 GOCART aerosols – GSI observation operators: • 15 GOCART species – Concentrations – Effective radius • CRTM internal optical parameters (from A da Silva and Dirceu Herdies) GEOS-5 overestimates dust Methodologies: Aerosol Radiance Contamination Dust Distribution for July 2008 event off West Coast of Africa (from A da Silva and Dirceu Herdies) Methodologies: Aerosol Radiance Contamination Temperature Analysis: DT = Taero - Tcontrol (from A da Silva and Dirceu Herdies) Methodologies: Aerosol Radiance Contamination Observation Count Residual Statistics Control Aero effects About 3% more AIRS observations are accepted (from A da Silva and Dirceu Herdies) Neutral impact to residual error statistics Methodologies: Lanczos Bi-Conjugate Gradient Objective: aid general formulation of WC-4dVar BiCG Double CG BiCG w/ ortho Double CG w/ ortho CG w/ ortho Lanczos CG Lanczos BiCG Remarks: - CG solves symmetric case - Double CG solves non-symmetric case - Double CG uses B-precond - Lanczos CG uses sqrt(B)-precond - BiCG solves non-symmetric case - Lanczos BiCG uses B-precond Changes entail: - add Bi-CG driver - mild glbsoi update - mild gsimod update - mild gsi_4dvar update Results highlight two aspects of CG: -Orthogonalization of gradients considerably improves convergence -Lanczos BiCG same as Lanczos CG, but former applies for non-symmetric case (from Amal El Akkraoui) Methodologies: Estimation of Q (model error) Q-st B-st Q-vp B-vp Q-t B-t Figure above shows normalized impact of observations within analysis window for SC and no-B WC. Plots show horizontal scales for B and prototype Q for stream function, velocity potential, and temperature at 45N obtained over a four-month sample of forecast full fields and tendencies, respectively. (from Banglin Zhang & Wei Gu) Closing Remarks • Completing comparison of SC and WC-4dVar in prototype GEOS-5 4dVar system. • Making progress in bringing GEOS-5 CubedSphere TLM and ADM to maturity. • Started working on hybrid ensemble components for GEOS-5 3d- and 4d-Var. Collaboration with NCEP is ongoing and fundamental for the success of these implementation. New Instruments: OMPS O3 (OSSE) OMPS – Ozone Mapping and Profiler Suite • Generate TRUTH • Simulate Radiances - - - GEOS-5.2.0 (MERRA tag) 1x1.25°L72 resolution Conventional data & satellite radiances impact meteorology Simple chemistry: O3 P&L in GCM MLS O3 profiles (215-0.1hPa) and OMI TC assimilated Hourly analysis output - - Interpolate TRUTH to OMPS/LP observation points to 1-km profile RT with pseudo-spherical atmosphere, multiple scattering, refraction, tangent shift, etc. Random surface reflectance, cloudtop height simulated and aerosol selected from SAGE-II database Validation • Assimilate Retrievals - - OMPS/LP data added to GSI in GEOS-5.6.1 The o3lev observer is used, same as for MLS QC flag for retrievals • Retrieve Profiles - - Rodgers’ Optimal Estimation Climatology as a-priori First retrieve cloud-top height, tangent height, surface reflectance and aerosol distributions Ozone profile retrievals • Apply Inst. Models - Instrument Simulator Model - Deconvolution Model - Consolidation Model