HFIP ENSEMBLE TEAM UPDATE Carolyn Reynolds (NRL) carolyn.reynolds@nrlmry.navy.mil Zoltan Toth (ESRL) zoltan.toth@noaa.gov Sim Aberson (HRD) Sim.Aberson@noaa.gov Tom Hamill (ESRL) tom.hamill@noaa.gov Jeff Whitaker (ESRL) Jeffrey.S.Whitaker@noaa.gov Fuqing Zhang (PSU) fzhang@psu.edu Yuejian Zhu (EMC) Yuejian.Zhu@noaa.gov Jun Du (EMC) Jun.Du@noaa.gov Mike Brennan (NHC) Michael.J.Brennan@noaa.gov Mrinal K Biswas (FSU) Mrinal.K.Biswas@noaa.gov Krish (FSU) tkrishnamurti@fsu.edu Develop more reliable and useful automated probabilistic numerical guidance for hurricane track, intensity, structure, rainfall, storm surge, and other associated weather elements through improved ensemble forecasting systems and improved postprocessing methods Work closely with HFIP Data Assimilation Team on development and use of ensemble-based data assimilation techniques for initializing ensemble predictions Work with Verification Team on developing and using ensemble/probabilistic measures Work with Products Team to develop ensemble/probabilistic products
NCEP/EMC Yuejian Zhu EMC ensemble team Contributors: Zhan Zhang and Jiayi Peng GEFS T574L64 work in progress Increased resolution (T126-T190) improves global ensembles Positive impact from bias correction and NAEFS for global ensembles Benefits seen from HWRF-GEFS over HWRF single run
Resolution makes difference for Typhoon Morakot Ini: 2009080600 T126 ensemble T190 ensemble Most members do not make right forecasts Ini: 2009080700 T126 ensemble T190 ensemble
Track forecast error for 2009 season (AL+EP+WP) 400 NCEPraw NCEPbc NAEFS 350 300 250 200 150 100 50 0 0 12 24 36 48 72 96 120 Cases 240 223 196 169 144 110 75 42 NAEFS is combined NCEP (NCEPbc) and CMC s (CMCbc) bias corrected ensemble and bias corrected GFS Contributed by Dr. Jiayi Peng (EMC/NCEP)
HWRF-GEFS Hurricane Ensemble System -- Uncertainty in Initial Large-Scale Flow Zhan Zhang HCTL: Control run, GFS input (T382L64), HWRF V3.2 (R2); ZEMN: Ensemble mean: GEFS input (T190L28), HWRF V3.2 (R2); 20+1 members; Two 2010 hurricanes: Alex (AL) and Celia (EP) Forecast skills are improved from 24h to 120h. The skill improvement increases with time.
Testing of the GFS EnKF for 2010 s TCs ESRL/PSD and GSD T. Hamill, J. Whitaker, P. Pegion and M. Fiorino
ENKF - ESRL/PSD and GSD ALL BASINS
ENKF - ESRL/PSD and GSD ATLANTIC
ENKF - ESRL/PSD and GSD ALL BASINS
ENKF - ESRL/PSD and GSD ATLANTIC
GSD/FAB REGIONAL ENSEMBLE CONFIGURATIONS Isidora Jankov et al. Task 3: Test/compare different ensemble configurations. Determine optimal ensemble strategy for regional ensembles, including initial perturbations and lateral boundary conditions. Attribute forecast uncertainty to initial condition vs. model related causes Test performance of various LAM dynamic cores and physical parameterizations Test cycling initial perturbation approach impact on TC forecasts Compare cycling results to regional EnKF results
GSD/FAB REGIONAL ENSEMBLE Very Preliminary Results (Bill 2009) LBCs from 6 different series of temporally consistent GEFS perturbed analyses WRF AWR, 6 km resolution 21-24 August 2009 CTRL PERT3
GSD/FAB - REGIONAL ENSEMBLE APPROACH Assume cloud of GEFS boundaries realistically capture large scale flow Test how well WRF AWR ensemble cloud can resolve finer scale storm development 6 km runs should resolve ~20km scales Collaborate with observing team Best estimate of truth on 20 km scale? Tweak physics / dynamics until LAM ensemble cloud captures proxy for reality Include stochastic physics, etc
GSD/FAB - REGIONAL ENSEMBLE ENSEMBLE CLOUD Temperature Xsect 12Z 21 Aug 2009 Wind (image) and Height Xsect Animation of different LAM runs
GSD/FAB STATISTICAL POST-PROCESSING / PRODUCTS Paula McCaslin & Ed Tollerud Objective: Test and compare different regional ensemble-based statistical post-processing methods Define & generate Storm Position Probability Distribution (SPPD) forecasts Generalized product related to strike probability Dressing of individual forecasts See look-alike product (instantaneous vs time integrated SPPD) Link up with products team; apply on various ensembles Decompose errors into positional & amplitude components Remove systematic positional & amplitude errors Use Bayesian principles Regime dependent corrections Weak vs. strong storms etc Statistical downscaling
SPPD User Interface Paula McCaslin NOAA/ GSD/ FAB
SPPD User Interface Paula McCaslin NOAA/ GSD/ FAB
SPPD Product Paula McCaslin NOAA/ GSD/ FAB
HRD/AOML Ensemble-based vortex-scale diagnostics T. Vukicevic, A. Aksoy and K. Sellwood Vortex scale properties of short-term forecasts are evaluated using the prior ensemble mean and covariances from cycling in HEDAS (Hurricane Ensemble Data Assimilation System). HEDAS is described in the data assimilation theme summary Ensemble-based vortex-scale diagnostic approach Each ensemble member and mean are projected onto cylindrical coordinate (R,Z, ), centered on the vortex center at each vertical level Forecast variables analyzed in (R,Z, ) coordinate system are tangential, radial and vertical wind components, temperature and humidity (Vt, Vr, W, T and Q, respectively) These variables are then azimuthally averaged resulting in wave number zero representation of the vortex short-term forecast Evolution of the prior and posterior mean and prior covariances over cycles in the data assimilation is then evaluated to asses properties of the short-term forecast errors
Diagnosing the modes of background forecast error Short-term dynamic adjustment in the forecast interferes with the analysis updates on vortex scale Mean prior and posterior primary circulation Error correlations Undesirable background forecast error mode: Systematic forecast spinning-down, opposing the spinning-up in the analysis Semi-undesirable background forecast error mode: Systematic forecast spinning-up is opposing the spinning-down in the analysis
PSU WRF-EnKF real-time ensemble assimilation and forecasting system for the 2010 Atlantic hurricane season Yonghui Weng and Fuqing Zhang Real-time 40.5- and 13.5-km WRF-EnKF ensemble analyses and 32-members of 4.5-km ensemble forecasts down to 4.5-km grid spacing twice per day from August 1 to September 30. Additional 4.5-km ensemble data assimilation in near realtime to assimilate airborne Doppler radar velocity if available, and produced additional subsequently ensemble forecast from these analysis and perturbations
PSU Real-time Production 1: EnKF Analyses Earl, 1002mb Earl, 996mb http://hfip.psu.edu/realtime/al2010/enkfd2cycle.html
PSU Real-time Production 3: Ensemble Forecast http://hfip.psu.edu/realtime/al2010/enkfd2cycle_tc.html
NRL C. Reynolds, S. Chen, J. Doyle, D. Hodyss, J. Goerss, T. Holt, J. McLay, J. Moskaitis, J. Ridout, C. Sampson, E. Serra, G. Carl COAMPS-TC part of FSU Multi-model Ensemble COAMPS-TC coupled ensemble capability in place NOGAPS ensemble parameter variations show small but significant improvement in TC track forecast. NOGAPS 20-member 42-km Ensemble run for 2010 season (diagnostics on-going)
NRL: COAMPS-TC ENSEMBLE Coupled Air-Ocean Forecasts of Hurricane Ike 29 member COAMPS-TC coupled ensemble forecasts of Hurricane IKE. Atmosphere: 81 km, 27 km Ocean: 27 km Perturbations: Initial conditions, LBCs, ocean initial state, model physics perturbations (cumulus, microphysics, PBL, surface fluxes). Moving atmospheric nest 2: 27-km t = 54-h Nest 1: 81-km t = 0-h Ocean nest 1: 27-km Ensemble Tracks (60 h) Best Track Coupled air-ocean ensembles provide probabilistic prediction of TC track, intensity, ocean response 27-km Nest
error (nm) NRL: NOGAPS Ensemble Research Ensembles with Parameter Variations 400 350 300 250 200 150 100 CTL PAR1 PAR2 Small improvements to TC track forecasts with inclusion of parameter variations (significant at 24, 72, and 96 h). 50 0 12 24 36 48 72 96 120 forecast time 103 93 77 69 52 30 17
NRL: NOGAPS Ensemble Research 20-mem High-res (42-km) Ensemble (Banded ET with SKEB): Homogenous All Basins Comparison: Sept-Oct 2010 Deterministic operational (blue); and ensemble mean (green) No significant difference in track errors. Both have slow bias (slightly larger for ensemble).
HFIP-THORPEX Ensemble Product Development Workshop: 20-21 April 2010, NCAR, Boulder, CO Hamill, Toth, Rappaport, DeMaria, Brown Co-hosted by HFIP, NOAA THORPEX, and the NCAR Tropical Cyclone Modeling Testbed (TCMT) Purpose: Review progress in ensemble prediction related to tropical cyclones (TCs) and the scientific issues in ensemble system development; Discuss new methods for displaying ensemble information that will aid forecasters Discuss what new uncertainty-based end products are a priority to develop. This workshop gave recommendations related HFIP research / development Deliverables Suggested modifications to ensemble and post-processing/product development work plans for HFIP program. A written report, suitable for publishing in BAMS and circulation among leadership of NOAA, NSF, NASA, ONR, and other funding agencies BAMS manuscript
FY11 Milestones from Strategic Plan Further develop and test methods for treating model error in ensembles, including stochastic parameterizations (S1 ESRL; in conjunction with data assimilation group). If warranted from experiments, operationally implement FY10 s stochastic convection/backscatter into operational global models (S2 NRL, ESRL, EMC) Determine whether fully coupled ocean model is required for hurricane global ensemble forecasts and whether simpler schemes (e.g., mixed-layer ocean model) are adequate replacements (ESRL/PSD pending funding) Evaluate two-way (atmos-ocean) coupled COAMPS-TC ensembles (S1 NRL) Evaluate of impact of improved initial perturbations (and implementation pending satisfactory results) into Navy global ensembles (S1 NRL) Initial test of regional ENKF for hurricane prediction (GSD-PSD, with DA team) Tools developed for attribution of initial vs. model errors in regional ensemble forecasts (GSD/FAB) Generalized Storm Position Probability Distribution (SPPD) forecasts (GSD/FAB)
Extra Slides
Extra Slides: NCEP
High resolution GEFS T574L64 (Plan) - For HFIP high resolution demonstration (2010) No results still working on the Oak Ridge machine High resolution global ensembles (NCEP/GEFS) T574L64 (~30km horizontal resolution) Initial analysis GSI T382L64 analysis (will be soon for T574L64 analysis) GFS model (latest GFS model 2010 version) ETR (ensemble transform with rescaling) Every 6 hours Cycling at T574L64 resolution Tuning for rescaling (based on T190L28 parameters) Upgrade analysis from T382L64 to T574L64 (if necessary) Integrations At Oak Ridge National Laboratory (ORNL) Use GFS model at T574L64 resolution (similar version as next implementation) 11 members (include control) Out to 168 hours With stochastic perturbations Near real time parallel Once per day from now (pending on the progress of set up) Experiments For summer 2008 Output Tracks for each members, ensemble mean, spread and etc..
Ensemble Track Forecasts, Hurricane Alex, 2010062606 35
Forecast skills are improved from 24h to 120h. The skill improvement increases with time.
Summary and Future Plan for HWRF-GEFS ensembles Hurricane track forecast skills are improved using HWRF-GEFS ensemble system (both Alex and Celia); Hurricane intensity forecast skills remained unchanged between ZEMN and HCTL; Physics-based perturbations: use different physics packages in HWRF; Combine GEFS perturbations with physicsbased perturbations; 37
Ensemble post processing and multi-model ensembles (NAEFS) - Jiayi Peng and Yuejian Zhu
Track forecast error for 2009 season (AL+EP+WP) 350 NCEPbc CMCbc GFS NAEFS 300 250 200 150 100 50 0 0 12 24 36 48 72 96 120 Cases 240 223 196 169 144 110 75 42 NAEFS is combined NCEP (NCEPbc) and CMC s (CMCbc) bias corrected ensemble and bias corrected GFS Contributed by Dr. Jiayi Peng (EMC/NCEP)
Summary and plan at EMC ensembles Global ensembles (partly set up and run experiments in 2009): High resolution ensembles at TACC machine (2009) T574 (23km), 10menbers (5 for gfs, 5 for fim) Improved ET initial perturbations (FY10/11 on going) Improved stochastic perturbations in physics (FY10 /11 on going) Meso-scale ensemble (under NEMS with various physics) working on progress: 20km resolution, 21 members ET initial perturbations (consistent with GEFS) Stochastic physics for convection possibly land surface perturbations (soil moisture, soil temperature etc.) 5-day integrations for case studies and possible experimental extension of operational SREF to 5-days for FY10/11 demo HWRF-GEFS hurricane ensemble system: Uncertainty in Initial Large-Scale Flow from GEFS; HWRF v3.2 and GEFS T126L28; 20+1 ensemble members; Preliminary results: improving tracks forecast, but not intensity Post-processing for storm related forecasts: Based on NCEP operational post processing in general for most variables Track forecast from NAEFS (multi-model) is best Future plan: Decompose gridded forecast errors into phase and amplitude component; Evaluate, then correct bias for phase before amplitude corrections;
GSD/FAB REGIONAL ENSEMBLE CONFIGURATIONS BACKUP SLIDES
Initial Perturbations for Cycling perturbations LAM forecast driven by global analysis Global Model Analysis interpolated on LAM grid Perturbations 00Z 06Z 12Z Forecast Time 42
24-hr Bill simulations by 5 members + control starting 21 Aug. 2009 at 00 UTC CTR EN1 EN2 EN3 EN4 EN5 985 985 988 988 988 985 00hr 991 990 993 992 992 989 06hr 998 996 998 998 998 996 12hr 1003 1002 1003 1004 1003 1002 18hr 1008 1007 1008 1008 1008 1007 24hr
Temperature Xsect 6-hr forecast Wind (image) and Height Xsect Animation of different LAM runs
Extra Slides: PSU
PSU WRF-EnKF realtime ensemble assimilation and forecasting system for the 2010 Atlantic hurricane season Yonghui Weng and Fuqing Zhang 60 Ensemble initialized at 1200UTC July 31 with GFS forecast 12hr lead time ensemble forecast EnKF Analyses With conventional observations Replace the ensemble mean with GFS analyses and Inherit the EnKF perturbations EnKF Analyses With conventional observations Ensemble forecast with 32 members EnKF Analyses With Airborne radar PSU WRF-EnKF realtime system work flow for 2010 Atlantic hurricane season
PSU Realtime Production 2: Ensemble Forecast 120h surface wind speed forecast initialized at 0000 UTC Aug 29 http://hfip.psu.edu/realtime/al2010/enkfd2cycle_forecast.html
NRL extra slides
Error (nm) Error (nm) Error (nm) NOGAPS Ensemble Research Ensemble Resolution vs. Number of Members ATL: Homogenous Sample Average Track Error WPAC: Homogenous Sample Average Track Error 350 300 250 200 150 100 50 0 12 EPAC:Homogenous 24 36 Sample 48 Average 72 Track Error 96 120 Forecast Hour 350 # fcsts: 356 320 281 245 182 129 86 G119 300 G159 G239 250 EPAC 200 G119 G159 G239 ATL 500 450 400 350 300 250 200 150 100 50 0 G119 G159 G239 WPAC 12 24 36 48 72 96 120 Forecast Hour # fcsts: 356 320 281 245 182 129 86 Biggest T119-T159 improvement seen in WPAC (submitted to MWR). 150 100 50 0 12 24 36 48 72 96 120 Forecast Hour
Slide on April Workshop