1 Part 2 CFS: Where It’s Going S. Lord, H-L Pan, S. Saha, D. Behringer, K. Mitchell And the NCEP (EMC-CPC CFSRR Team)

2 Overview Current (CFS-v1) description and status CFS Reanalysis and Reforecast (CFSRR  CFS-v2) –Atmosphere –Ocean –Land surface –Sea ice Future development (CFS-v3) –Coupled A-O-L-S system –Long term Reanalysis strategy Possibilities for Multi-Model Ensembles (MMEs) Weather-Climate forecasting

3 Seasonal to Interannual Prediction at NCEP Operational System since August 2004 (CFS-v1) Climate Forecast System (CFS) Ocean Model MOMv3 quasi-global 1 o x1 o (1/3 o in tropics) 40 levels Atmospheric Model GFS (2003) T62 (~200 km) 64 levels GODAS (2003) 3DVAR XBT TAO Triton Pirata Argo Salinity (syn.) TOPEX/Jason-1 Reanalysis-2 3DVAR T62L28 OIv2 SST Levitus SSS clim. Ocean reanalysis (1980-present) provides initial conditions for retrospective CFS forecasts used for calibration and research Stand-alone version with a 14-day lag updated routinely Daily Coupling “Weather & Climate” Model Funded by NCPO/OCO

4 Number of Temperature Observations per Month as a Function of Depth D. Behringer

5 1.High resolution data assimilation –Produces better initial conditions for operational hindcasts and forecasts (e.g. MJO) –Enables new products for the monthly forecast system –Enables additional hindcast research 2.Coupled data assimilation –Reduces “coupling shock” –Improves spin up character of the forecasts 3.Consistent analysis-reanalysis and forecast-reforecast for –Improved calibration and skill estimates 4.Provide basis for a future coupled A-O-L-S forecast system running operationally at NCEP (1 day to 1 year) –(currently in parallel testing for “GFS” 1-14 day prediction) CFS-v2 Highlights Funded by NCPO/CDEP

6 CFSRR Components Reanalysis –31-year period ( and continued in NCEP ops) –Atmosphere –Ocean –Land –Seaice –Coupled system (A-O-L-S) provides background for analysis –Produces consistent initial conditions for climate and weather forecasts Reforecast –28-year period ( and continued in NCEP ops ) –Provides stable calibration and skill estimates for new operational seasonal system Includes upgrades for A-O-L-S developed since CFS originally implemented in 2004 –Upgrades developed and tested for both climate and weather prediction –“Unified weather-climate” strategy (1 day to 1 year)

7 CFSRR Component Upgrades ComponentOps CFS2010 CFS Atmosphere 1995 (R2) model 200 km/28 sigma levels 2008 model (upgrades to all physics) 38 km/64 sigma-pressure levels Enthalpy-based thermodynamics Variable CO2 (historical data, future scenarios) R2 analysis Satellite retrievals GSI with simplified 4d-var (FOTO) Radiances with bias-corrected spinup Ocean MOM-3 60N – 65 S 1/3 x 1 deg. MOM-4 Global domain ¼ x ½ deg. Coupled sea ice forecast model Ocean data assim. 750 m depth2000 m Land No separate land property analysis Global Land Data Assim. Sys (GLDAS) driven by observed precipitation 1995 land model (2 levels)2008 Noah model Sea ice Daily analysisDaily hires analysis Coupling NoneFully coupled background forecast (same as free forecast)

8 00Z GDAS06Z GDAS18Z GSI 0Z GODAS 00Z GSI One-day schematic of four 6-hourly cycles of CFSRR Global Reanalysis: 6Z GODAS12Z GODAS0Z GODAS 12Z GDAS 18Z GODAS Atmospheric Analysis Ocean Analysis 12Z GLDAS6Z GLDAS18Z GLDAS0Z GLDAS Land Analysis Time S. Saha and S. Moorthi

9 Testing with CMIP Runs (variable CO2) OBS is CPC Analysis (Fan and van den Dool, 2008) CTRL is CMIP run with 1988 CO2 settings (no variations in CO2, current operations) CO2 run is the ensemble mean of 3 NCEP CFS runs in CMIP mode –realistic CO2 and aerosols in both troposphere and stratosphere Processing: 25-month running mean applied to the time series of anomalies (deviations from their own climatologies)

10 CFSRR at NCEP GODAS 3DVAR Ocean Model MOMv4 fully global 1/2 o x1/2 o (1/4 o in tropics) 40 levels Atmospheric Model GFS (2007) T levels Land ModelIce Model LDAS GDAS GSI 6hr 24h r 6hr Ice Ext 6hr Climate Forecast System V2

11 Future Development What’s going on and what’s needed –Land surface –Ocean & Sea ice –Atmosphere

12 Noah LSM replaces OSU LSM in new CFS Noah LSM –4 soil layers (10, 30, 60, 100 cm) –Frozen soil physics included –Surface fluxes weighted by snow cover fraction –Improved seasonal cycle of vegetation cover –Spatially varying root depth –Runoff and infiltration account for sub-grid variability in precipitation & soil moisture –Improved soil & snow thermal conductivity –Higher canopy resistance –More OSU LSM –2 soil layers (10, 190 cm) –No frozen soil physics –Surface fluxes not weighted by snow fraction –Vegetation fraction never less than 50 percent –Spatially constant root depth –Runoff & infiltration do not account for subgrid variability of precipitation & soil moisture –Poor soil and snow thermal conductivity, especially for thin snowpack and moist soils Noah LSM replaced OSU LSM in operational NCEP medium-range Global Forecast System (GFS) in late May 2005 Some Noah LSM upgrades & assessments were result of collaborations with CPPA PIs Funded by NCPO/CPPA K. Mitchell

13 CFSRR Reanalysis Land Component: Global Land Data Assimilation System (GLDAS) Applies same Noah LSM as in new CFS Uses same native grid (T382 Gaussian) as CFSRR atmospheric analysis Applies CFSRR atmospheric analysis forcing (except for precip) –hourly from previous 24-hours of atmospheric analysis –Precipitation forcing is from CPC analyses of observed precipitation Model precipitation is blended in only at very high latitudes GLDAS daily update of the CFSRR reanalysis soil moisture states –Reprocesses last 6-7 days to capture and apply most recent CPC precipitation analyses Realtime GLDAS configuration will match reanalysis configuration –To sustain the relevance of the climatology of the retrospective reanalysis Applies LIS: uses the computational infrastructure of the NASA Land Information System (LIS), which is highly parallelized

14 LIS Capabilities Flexible choice of 7 different land models –Includes Noah LSM used operationally by NCEP and AFWA Flexible domain and grid choice –Global: such as NCEP global model Gaussian grid –Regional: including very high resolution (~.1-1 km) Data Assimilation –Based on Kalman Filter approaches High performance parallel computing –Scales efficiently across multiple CPUs Interoperable and portable –Executes on several computational platforms –NCEP and AFWA computers included Being coupled to NWP & CRTM radiative transfer models –Coupling to WRF model has been demonstrated –Coupling to NCEP global GFS model is under development –Coupling to JCSDA CRTM radiative transfer model is nearing completion Next-gen AFWA AGRMET model will utilize LIS with Noah NCEP’s Global Land Data Assimilation utilizes LIS K. Mitchell, C. Peters-Lidard

15 Impact of Noah vs. OSU Land Models and GLDAS Initial Land States in 25-years of CFS Summer & Winter Reforecasts: Lessons Learned Land surface model (LSM) for CFS forecast must be same as for supporting land data assimilation system (LDAS) Impact of land surface upgrade on CFS seasonal precipitation forecast skill for is positive (but modest) –Significant only for summer season in neutral ENSO years (and then only small positive impact) –Essentially neutral impact for winter season and non-neutral ENSO summers Differences in CFS precipitation skill over CONUS between neutral and non-neutral ENSO years exceeds skill differences between two different land configurations for same sample of years –Indicates that impact of SST anomaly is substantially greater than impact of land surface configuration

Land Surface Model Development 1 - Unify all NCEP model land components to use MODIS-based hi-res global land use with IGBP classes 2 - Improve global fields of land surface characteristics (vegetation cover, albedo, emissivity) using satellite data (with Joint Center for Satellite Data Assimilation) 3 - Enhance land surface subgrid-variability with high-resolution sub-grid tiles 4 - Increase number of soil layers (from 4 to about 10) 5 - Introduce dynamic seasonality of vegetation (to replace pre-specified seasonal cycle) 6 - Improve hydrology including addition of groundwater 7 - Add multi-layer treatment to snowpack physics 8 - Introduce carbon fluxes Items 5-8 are being transitioned from the CPPA-funded work of PI Prof Z.-L. Yang and Dr. G.-Y. Niu of U.Texas/Austin K. Mitchell

17 Operational in 2010 MOMv4 (1/2 o x 1/2 o, 1/4 o in the tropics, 40 levels) Updated 3DVAR assimilation scheme –Temperature profiles (XBT, Argo, TAO, TRITON, PIRATA) –Synthetic salinity profiles derived from seasonal T-S relationship –TOPEX/Jason-1 Altimetry –Data window is asymmetrical extending from 10-days before the analysis date –Surface temperature relaxation to (or assimilation of) Reynolds new daily, 1/4 o OIv2 SST –Surface salinity relaxation Levitus climatological SSS –Coupled atmosphere-ocean background Current stand-alone operational GODAS will be upgraded in 2009 to the higher resolution MOMv4 and be available for comparison with the coupled version –Updated with new techniques and observations GODAS in the CFSRR D. Behringer

18 Assimilating Argo Salinity ADCP GODAS GODAS-A/S In the east, assimilating Argo salinity reduces the bias at the surface and sharpens the profile below the thermocline at 110 o W. In the west, assimilating Argo salinity corrects the bias at the surface and the depth of the undercurrent core and captures the complex structure at 165 o E. Comparison with independent ADCP currents. D. Behringer

GODAS Activities Complete CFSRR –Evaluate ODA results Add ARGO salinity Improve climatological T-S relationships and synthetic salinity formulation ENVISAT data? Improve use of surface observations –Vertical correlations (mixed layer) Situation-dependent error covariances (recursive filter formulation) Investigate advanced ODA techniques –Experimental Ensemble Data Assimilation system (with GFDL) –Reduced Kalman filtering (with JPL) –Improved observation representativeness errors (with Bob Miller, OSU- JCSDA) Impact of the GODAS mixed layer analysis on subseasonal forecasting with the CFS. Augustin Vintzileos (EMC) D. Behringer

20 Comparison of GODAS/KF and GODAS/3DVAR with TAO temperature and zonal velocity anomalies R e = [model explained variance] / [data variance] 20 o CDynHtSSTU 3DVAR - A KF 3DVAR - B KF SST 20 o CDynHtU in collaboration with I Fukumori (JPL) For points toward the top (GKF) and toward the right (G3DV) the models are closer to the data. For points above (below) the diagonal GKF (G3DV) is closer to the data.

21 Sea Ice Analysis from CFSRR R. Grumbine

22 Atmospheric Model Improve CFS climatology and predictive skill with improved physical parameterizations –Deep and/or shallow convection –Cloud/radiation/aerosol interaction and feedback –Boundary layer processes –Orographic forcing –Gravity wave drag –Stochastic forcing –Cryosphere

23 Shallow Cloud Development H.-L. Pan and J. Han Use a bulk mass-flux parameterization Based on the simplified Arakawa-Shubert (SAS) deep convection scheme, which is being operationally used in the NCEP GFS model Separation of deep and shallow convection is determined by cloud depth (currently 150 mb) Main difference between deep and shallow convection is specification of entrainment and detrainment rates Only precipitating updraft in shallow convection scheme is considered; downdraft is ignored

24 Siebesma & Cuijpers (1995, JAS) Siebesma et al. (2003, JAS) LES studies Development based on LES studies

25 ISCCP Control Revised PBL & new shallow convection Cloud cover improved Combined Impact of Revised PBL & New Shallow Convection For CFS J. Han

26 Revised PBL + New shallow (Winter 2007) NH(20N-80N)SH(20S-80S) 500 hPa Height Anomaly Correlation Skill scores are better (1)

hrs hrs hrs CONUS Precipitation skill score Winter 2007 Skill scores are better (2)

28 Revised PBL + New shallow (Summer 2005) NH(20N-80N)SH(20S-80S) 500 hPa Height Anomaly Correlation Skill scores are better (3)

hrs hrs hrs CONUS Precipitation skill score Summer 2005 Skill scores are possibly better (4)

30 ENSO Signal Observed SST Anomaly Nino 3.4 OIV2Control SST Anomaly 50 year CMIP Run ENSO too weak (early) Too strong later RESULT: no implementation for Weather or climate

31 Observed DSWR from Visiting Scientist (Mechoso – UCLA, CPPA sponsored through VOCALS) Downward Shortwave Radiation at Ground 2S-2N Annual Mean 50 Year Run Observed Year 1-20Year 21-50ControlYear 1-20ExperimentYear Clouds Too Thick in SE Pacific (DSWR too small) Can be Improved With Shallow-Deep Cloud Tuning

32 Phase (local time) of Maximum Precipitation (24-hour cycle) Five-member ensembles driven by Climatological SST forcing ( avg) Myong-In Lee and Sieg Schubert (NASA/GMAO)

33 Impact of Diurnal SST (Xu Li)

34 T254 T126 T62 GDAS CDAS-2 RMS Error Growth Pattern Correlation Resolution does not affect skill. Forecasts initialized by GDAS are better (a gain of ~3-5 days). Time evolution of mean energy at wave numbers when CFS is initialized by R2 (red) or by GDAS (blue). drift Tropical Intraseasonal Forecasts (MJO) A. Vintzileos

35 Ongoing Reanalysis Project CFS will be upgraded every ~7 years –New forecast system Upgrades from operations New techniques Higher resolution analysis Aerosol and trace gas analysis Carbon cycle Hydrology, ground water, etc. –New observations from data mining –Satellite data treatment (e.g. bias correction) Evolution to Integrated Earth System Analysis Ongoing work to incorporate these improvements –Preparation for Reanalysis production phase –All additions carefully tested

36 Proposed Concept of Operations Production Phase 2-3 years Development Phase 3-4 years

37 Future Model Component Upgrades Component2010 CFSPossible Upgrades Atmosphere - AER RRTM shortwave & longwave radiation - Variable CO2 & aerosols - Maximum random cloud overlap Enthalpy-based thermodynamics - Fractional cloudiness (impacts surface solar flux) - Possible neural network emulation for radiation (trained on hindcasts) - Sigma-pressure-theta hybrid - Prognostic cloud water - Non-local PBL - Simplified Arakawa-Schubert conv. - Ferrier microphysics (impacts radiation and precipitation type) - Shallow convection (mass flux) - Convective gravity wave - Conservative, positive definite tracer advection Land - Global Land Data Assim. Sys (GLDAS) driven by observed precipitation - Dynamic vegetation (impacts drought) - Groundwater (impacts soil wetness) Ocean - MOM-4-Ocean ensemble (HYCOM + MOM ?) -Salinity assimilation - Situation-dependent background errors and other advanced techniques Comprehensive Testing in Weather and Climate Modes Daily data assimilation and 15 day forecasts LDAS for balanced land states CMIP runs (> 50 years) Sample seasonal runs (May & October)

38 Multi-Model Ensemble Strategy International MME (IMME) with EUROSIP is under negotiation –Operational delivery –Consolidated products –Use for official duty only Full set of hindcasts required for bias correction and skill masking National MME –COLA is generating hindcasts for NCAR system –Issues are developing concept of operations (how partners will participate) identifying metrics for value added (e.g. consolidation) building computing resources (particularly for reforecasts) into computer acquisitions

39 IMME Status (1) Goal: produce operational ensemble products from CFS and EUROSIP seasonal climate products EUROSIP –ECMWF –Met Office –Meteo France Prospectus has been submitted to EUROSIP Counsel –Covers Licensing Commercial interest and revenue sharing –Consistent with EUROSIP general provisions Formal Memorandum of Understanding will be drafted

40 IMME Status (2) Some tenets of a potential agreement –E-partners and NCEP will be free to process individual forecasts into combined IMME products with their own procedures –NCEP will distribute its combined IMME product to its internal users for official duty use in time to meet NCEP forecast schedules –NCEP will distribute its combined products to the E-Partners as soon as possible each month, using ECMWF as the distributing agent –NCEP and E-partners will coordinate distribution of IMME products to their users on a regular monthly schedule –Product delivery will not compromise any organization’s operational delivery schedules and commitments –NCEP wishes to join the EUROSIP Steering Group as a non- voting member and will participate in future meetings

41 Weather-Climate Forecasting

42 GDAS GFS anal NAM anal CFS RTOFS SREF NAM AQ GFS HUR RDAS Current (2007) GENS/NAEFS Current NCEP Production Suite Weather, Ocean, Land & Climate Forecast Systems

43 Global Model Suite Daily to S/I Forecasts All forecasts are Atmosphere-Land-Ocean coupled All systems are ensemble-based except daily, high-resolution run All forecasts initialized with LDAS, GODAS, GSI from GFS initial conditions Physics and dynamics packages may vary –Anticipated that the weekly forecast will have most rapid implementations and code changes, seasonal configuration may be one (or at most two) versions behind weekly Forecast Product Membership refresh period Runs/dayNumber of members per refresh period Horizontal resolution (ratio, current value) Forecast Length Initialization technique Computing Resource ratio Daily-hires4x/day411.0, T38215 daysGSI1.0 Weeklydaily80 0.5, T17015 daysET breeding2.5 Monthlyweekly8560.5, T17060 days??1.0 Seasonalmonthly , T1261 yearLagged analysis 4x daily 0.44

44 CFS & MFS CFS MFS Regional Rap Refresh Global Hydro Next Generation Prototype AQ WAV GFS SREF Reforecast NAM RTOFS CFS & MFS & GODAS AQHydro / NIDIS/FF GENS/NAEFS HUR HENS NCEP Production Suite Weather, Ocean, Land & Climate Forecast Systems GDAS RDAS

45 Summary CFSRR  CFS-v2 –High resolution reanalysis –CO2 trend –Upgrades models and data assimilation –Foundation for coupled “earth-system” reanalysis Beginning scientific development of CFS-v3 –Fully coupled A-O-L-S system for IESA –Advanced data assimilation techniques Building a MME system with International and US contributions Focusing on Weather-Climate forecasting –1 day to 3 years

46 Thanks Questions?

47 Comparison of GODAS/M4 and GODAS/M3 with TAO temperature and zonal velocity In the thermocline both GM4 and GM3 are warm at 140w, while GM4 is warm and GM3 is cold at 110w. The undercurrent is stronger than observed in GM4 and weaker in GM3. The vertical structure at 165e is better in GM4 than in GM3.

48 Land Information System (LIS) NOAA-NASA-USAF collaboration –K. Mitchell (NOAA) –C. Peters-Lidard (NASA) –J. Eylander (USAF) LIS hosts –Land surface models –Land surface data assimilation and provides –Regional or global land surface conditions for use in Coupled NWP models Stand-alone land surface applications K. Mitchell, C. Peters-Lidard

49 Science Plan for the CFS (II) Most effective way to improve the CFS (climate) GFS/CFS (weather) as one package We want to improve weather and climate forecasts by making physically based improvements to the atmospheric model parameterization packages. We have been successful when we apply rigorous tests to physically based parameterization improvements to both weather and climate models and want to continue along this way.

50 Science plan for the CFS (III) Deep and/or shallow convection These processes transport sub-grid scale heat and moisture vertically, which is especially important for climate prediction. Boundary layer processes As the CFS is a coupled model, the boundary layer is critical for communication of the ocean and land conditions with the atmosphere. Cloud/radiation/aerosol interaction and feedback Clouds and aerosols modulate the sources and sinks of the thermal energy in to the earth system. This interaction is crucial on climate time scales. Orographic forcing Orography determines many climate variables through form-drag, mountain blocking, and land/sea contrast.

51 Science Plan for the CFS (IV) Gravity wave drag Gravity waves generated by the sub-grid scale orography and/or cumulus convection transport wave energy from the troposphere to the stratosphere and mesosphere and thus control the climate of those regions. Stochastic forcing Stochastic forcing is not in the CFS at this time, but is important for parameterizing random, unresolved physical forcing. Cryosphere The cryosphere (glaciers, frozen land, sea ice) plays a crucial role in determining the earth's climate. Modeling of sea-ice and its interaction with the ocean and atmosphere, and modeling frozen land and its interaction with the atmosphere are all important to climate.

52 Science Plan for the CFS (V) Testing procedures are key to the road to making model implementations While transition to operation for MMEs requires only seasonal hindcasts to be evaluated, it is done because we expect the team maintaining the MME models to do their own rigorous tests. Tests in data assimilation modes and evaluated with forecasts are crucial for weather forecasts. Tests in multi-year coupled simulations and seasonal hindcasts are crucial for climate forecasts CTB computer resource is not sufficient and NCEP computer must be used when full-scale testing is needed

53 Gaps Insufficient EMC staff to collaborate with external investigators, train their staff (often post-docs) on use of the CFS, and develop new parameterization codes suitable for the CFS for the broad spectrum of possible areas listed above (O2R); Insufficient computing resources for experimentation and transition changes to the CFS; Insufficient EMC and NCEP Central Operations (NCO) staff to support the R2O (implementation) process; Insufficient knowledge within the research community about the tests needed to complete an implementation

54 We built a new shallow convection scheme a few years ago Use a bulk mass-flux parameterization Based on the simplified Arakawa-Shubert (SAS) deep convection scheme, which is being operationally used in the NCEP GFS model Separation of deep and shallow convection is determined by cloud depth (currently 150 mb) Main difference between deep and shallow convection is specification of entrainment and detrainment rates Only precipitating updraft in shallow convection scheme is considered; downdraft is ignored

55 Siebesma & Cuijpers (1995, JAS) Siebesma et al. (2003, JAS) LES studies We build it based on LES studies

56 Cloud water cross-section looks better

57 PBL & Low clouds combined (CFS run) ISCCP Control Revised PBL & new shallow convection Cloud cover looks better

58 Revised PBL + New shallow (Winter, 2007) NH(20N-80N)SH(20S-80S) 500 hPa Height Anomaly Correlation Skill scores were better

hrs hrs hrs Precipitation skill score over US continent Skill scores were better

60 Revised PBL + New shallow (Summer, 2005) NH(20N-80N)SH(20S-80S) 500 hPa Height Anomaly Correlation Skill scores were better

hrs hrs hrs Precipitation skill score over US continent Skill scores were slightly better

62 NINO3.4 OIV2 Observed ENSO signal

63 NINO3.4 set22 Multi-year simulation of the control looks ok

64 NINO3.4 set28b The test version showed too weak ENSO in early years and too strong ENSO in later years. RESULTS : no implementation

65 With a VOCALS grant from CPPA, Mechoso worked with us to examine these runs. This is the downward short wave radiation reaching ground for the control Srb2 is observation

66 There is too much radiation reaching ground for the new package over western Pacific but too little over central Pacific. More changes will have to be made.

67 Cloud water cross-section looks better

68 Climate Requirements for NCEP’s Next Operational System (2011) ApplicationOperational SystemComputing X factor Requirement Generator Seasonal-Monthly Climate CFS32Climate Prediction Center Monthly fcst system10 + “ GLDAS, NLDAS2.5*NIDIS, CPC Reforecast9* “ Ratio of ops:R2O computing –Currently 1:1.3 –Requesting 2011: 1: : 1: : 1:4.0 + Extension of Week2 system * New system Climate (and other) computing requirements total a factor of 3X in additional funding (above Moore’s Law – constant $$ capability)

69 NOAA Computing Resources and Operational Requirements for Climate Forecasting at NCEP Research –Including CFSRR Operations

70 Climate R&D Computing for Week2 to S/I June New Power6 system for CTB, CDEV, JCSDA, MTB (same % as previous) CFSRR will use all Power5 system Enables CFSRR to execute ¾ of required production rate