NCEP’s Global Forecasting System (GFS) and other Recent Developments at the NCEP Environmental Modeling Center Stephen Lord (Director) and EMC Staff 1 2 Currently being upgraded again - major changes to allow easier upgrades in future 3 Overview • • • • • • Global weather NASA-NOAA-DOD JCSDA Climate Mesoscale weather Ocean Hurricanes 4 Global Weather • Global Forecast System (GFS) • Global Ensemble Forecast System (GEFS) – Extended T126 horizontal resolution after 180 hours – Initial perturbations (breeding cycle): • Fixed bugs for calculating re-scaling factors • Use 6-h breeding instead of 24-h breeding • Adjust mask to tune initial perturbation – Tropical storm relocation • Add TS relocation scheme to ensemble initial perturbation (5% of total storm size). 5 Global Forecast System (GFS) model Configuration GFS Spectral Sigma T382/L64 (0-180 h) T190/L64 (180-364 h) Global GFS Simple Cloud Modified Arakawa/ Schubert GFS Chou (SW), GFDL (LW) Ferrier JCSDA SST NESDIS/USAF NOAH Land Model 6-Layer Model Burk and Thompson 6 GFS - List of Upgrades • Global Forecasgt System (GFS) – Increase resolution from T254 (55 km) to T382 (35 km) • Old: T254/L64 (0-84 h) T170/L42 (84-180h, T126/L28 to 384h • New: T382/L64 (0-180 h) T190/L64 (180-364 h) – Modified vertical diffusion – Enhanced mountain blocking – New sea ice model • Fractional sea ice & leads • Impacts surface fluxes – New code structure • Increased computational efficiency • ESMF compatible superstructure • “Hybrid (sigma-pressure) ready” 7 GFS - List of Upgrades (cont) • Model (cont) – Upgrade to Noah Land Surface Model • 2-4 soil layers • Reduction of early bias in snow pack depletion • Improved treatment of – Frozen soil – Ground heat flux – Energy and water balance at surface • Reformulated infiltration and runoff functions • Upgraded vegetation fraction (NESDIS) • Improved, plug-compatible, code structure 8 GFS - List of Upgrades (cont) • Analysis – 3D-VAR – Increase resolution to T382 – Surface emissivity model for snow and ice (JCSDA) • 3 X data used in SH polar latitudes • 1.3 X in NH polar latitudes – – – – AQUA AIRS and AMSU-A (new data) AIRS (new data) Upgraded thinning algorithm for radiances QC algorithm for clouds 9 Performance Results - Winter AC +2% RMS - 8% Consistent day-to-day performance 10 Performance Results – Summer AC +3% RMS - 8% Consistent day-to-day performance 11 Global Weather • Global Forecast System (GFS) • Global Ensemble Forecast System (GEFS) – Extended T126 horizontal resolution after 180 hours – Initial perturbations (breeding cycle): • Fixed bugs for calculating re-scaling factors • Use 6-h breeding instead of 24-h breeding • Adjust mask to tune initial perturbation – Tropical storm relocation • Add TS relocation scheme to ensemble initial perturbation (5% of total storm size). 12 Increasing spread for week-two forecast Deterministic Runs Ensemble Runs Black-opr Red-exp 13 SH RMS and spread Improved outlier SH ROC SH RPSS Improved skill for short, extended-range forecast 14 Hurricane Track Plots (case 1) Frances (08/28) With relocation Without relocation 15 Track errors and spreads 2004 Atlantic Basin (8/23-10/1) opr-errors 400 exp-errors opr-spread exp-spread From Timothy Marchok (GFDL) 350 300 250 200 150 100 Reduced mean track errors and spreads 50 0 24h 48h 72h 96h 120h 16 Overview • • • • • • Global weather NASA-NOAA-DOD JCSDA Climate Mesoscale weather Ocean Hurricanes 17 NCEP Data Assimilation system is moving form Spectral Statistical Interpolation (SSI) to Grid-point Statistical Interpolation (GSI) Continuing with 3D-VAR for now, due to product delivery timetable constraints NCEP Global Forecast System must begin delivering products to users within about 3 hours of radiosonde observations Also doing tests with alternative forms of 4D-VAR 18 Monthly Mean TBges-TBobs AMSU-A1&A2 NOAA15 A1 A2 A3 A5 New A1 A2 current 19 GPSRO Data Assimilation Development (L. Cucurull) • Current and short-term work: – Implementation of the local Bending Angle Forward Operator (BAFO) in the GSI. – Examination of representativeness error. • If realistic errors (for refractivity and bending angle along the vertical) are not available in time for impact studies, we will use simulated errors. – Compare impact studies between BAFO and RFO and select the Forward Operator for COSMIC. – Develop (real-time) Monitoring System (O-B) for calibrating/testing the RO 20 Assimilation of RO + other data RO locations 22 Getting COSMIC data to Weather Centers NCEP I n p u t D a t a C D A A C BUFR Files WMO standard 1 file / sounding N E S D I S ECMWF GTS CWB UKMO JMA Canada Met. This system is currently under development by UCAR, NESDIS, & UKMO 23 New JCSDA SATELLITE SST Retrieval Method: • Developed as alternative SST retrieval method – based on a physical retrieval (variational) algorithm which runs within the structure of the GDAS (Derber and Xu Li). – Cost function minimizes the increment between; • Observed radiances and simulated radiances, and • Analyzed SST and its first guess • Requires radiative transfer model to simulate Brightness Temperatures for each channel using – SST first guess (previous analysis) – Air Temperature (GDAS analysis) – Water vapor mixing ratio (GDAS analysis) 24 Smoother anomalies (less noise) Smoother anomalies (less noise) 25 Daily Analysis Difference RTG_SST-HR Operational Reduced daily noise 26 Data Assimilation Status • Gridpoint Statistical Analysis (GSI) – NCEP’s next generation system • Evolutionary combination of the global SSI analysis system and the regional ETA 3DVAR • Application to both global and regional analysis • Strong heritage to satellite, radar, profiler, surface data – Background error defined in grid space instead of spectral space • Allows use of situation dependent background errors • Will accept ensemble information – Improved balance condition • Adiabatic dynamics model • Capable of simplified 4-D Var – Improved and modernized code • F90/95 structures and utilities • Increased scalability of code • Efficiency – Redesigned data distribution – Some OpenMP • Better documentation • Less dependency on IBM – Community support intended but not resourced • Currently 15 registered groups (46 users) using GSI code • NASA/GMAO major group using code and to date they have provided the most updates from external users 27 Overview • • • • • • Global weather NASA-NOAA-DOD JCSDA Climate Mesoscale weather Ocean Hurricanes 28 NCEP Performance Comparison Seasonal Forecasts Raw Nino3.4 SST Correlation Skill Annual Mean 1981-2001 NCEP CFS wrt OIv2 1971-2000 climatology 100 CA (Statistical) CFS ECM MFR DEMETER Anomaly Correlation [ % ] 80 60 MPI UKM ING 40 LOD CER CA 20 0 1 2 3 Forecast Lead [ months ] 4 5 29 Overview • • • • • • Global weather NASA-NOAA-DOD JCSDA Climate Mesoscale weather Ocean Hurricanes 34 Katrina Multi-model Consensus CONU: GFS UKMO NGPS GFDL GFDN 35 36 Examples of GFS performance Compared with: 1 - Other Global models 2 - NCEP’s North American Mesoscale models 37 Human forecaster { GFS Ensembles 38 Precipitation Forecast Comparisons (Threat & Bias) Model groups International (1)* Period 10/2004 – 9/2005 Forecast f24-f48 North American+ 10/2004 – 9/2005 f24-f48 North American International (2)* International (2)* North American 10/2004 – 9/2005 10/2004 – 6/2005 7/2005 – 9/2005 7/2005 – 9/2005 f24 f24-f72 f24-f72 f24-f48 * International global (1): DWD, ECMWF, JMA, UK, USA International global (2): CMC, DWD, ECMWF, JMA, USA + North American: USA (global), USA (NAM), CMC (global), CMC (regional) 39 Precipitation Verification at NCEP/EMC Precipitation forecasts from various operational and parallel (experimental) models are verified over ConUS and its 14 sub-regions: • Daily (12Z-12Z) verification against daily gauge analysis (7,000-8,000 gauges) • 3-hourly verification against NCEP Stage II multi-sensor hourly analysis • Verification for international models for 24 h accumulated amounts over ConUS domain Statistics on the number of forecast/correctly forecast/observed forecast points (FHO) are collected for various precipitation thresholds. Dozens of scores can be computed from the FHO database, e.g. equitable threat, bias, probability of detection, odds ratio. 40 Global Models – Annual 1 October 2004 – 30 September 2005 24 – 48 h Forecasts 41 Global Models – Cold season 1 October 2004 – 30 June 2005 24 – 72 h Forecasts 42 Global Models – Warm season 1 July 2005 – 30 September 2005 24 – 72 h Forecasts 43 North American Models - Annual 1 October 2004 – 30 September 2005 24 – 48 h Forecasts 44 North American Models – Warm season 1 July 2005 – 30 September 2005 24 -48 h Forecasts 45 All operational data available on web without any restrictions for use 46