Recent Upgrades and Plans for the
NOAA/NCEP Short Range Ensemble
Forecast (SREF) System
Jeff McQueen, Jun Du, Binbin Zhou, Geoff Manikin,
Brad Ferrier and Geoff DiMego
Tuesday, August 1, 2017
SREF Team
• System Integration/Operations:
• Physics Diversity Configuration:
• Product Generation/Visualization:
• Standard Suite:
• Aviation, Energy:
• Severe Weather:
Jun Du
B. Ferrier
Binbin Zhou, Jun Du
Binbin Zhou
G. Manikin, D. Bright
• Verification:
– Model to Observations (Det/Prob):
– Model to analysis (Det/Prob):
– Case Studies:
H. Chuang
B. Zhou
G. Manikin, R. Grumm
• Calibration:
– Bias Correction:
– Bayesian Model Averaging:
• High Res Ensembles (WRF):
•
•
Ensemble Transforms (Future):
Training:
J. Du, B. Coi
Mark Raulston
G. DiMego, D. Jovic,
E. Rogers, H. Chuang
M. Wei, Z. Toth
B. Bua
Outline
• Improved SR-Ensemble Prediction Systems
– NCEP Short Range Ensemble Forecasts (SREF)
– High Resolution Window Weather Reseach and Forecast
System (WRF) Ensemble
• Improved Deterministic and Probabilistic Products
– Higher Fidelity  Capture smaller scale features
– Improved Accuracy
– Improved probabilistic information to help quantify forecast
uncertainties
– Bias Correction and Bayesian Model Averaging
– Visualization
– Verification
Ensemble Modeling System Goals
• Improved probabilistic products for NWS mission
forecasts (Severe storms, Aviation, Hydromet, ocean,
tropical, Energy, Dispersion)
• Quantify Uncertainty for Each Forecast Run
– High Confidence= good agreement between forecasts?
• Improved Spread-Skill relationship Information
– System variance ~ System Mean Squared Error
– Less clustering among ensemble members(more spread)
• Improved or similar skill as determined from
ensemble mean and probabilistic skill scores for 1-3
day forecasts (Skill scores, Sharpness of
probabilistic forecast) :
– Temperatures, winds, moisture
– Precipitation
– Upper-level winds, heights
Recent SREF Improvements
•
•
Increased Resolution
•
48 km to 32 km horizontal resolution
•
Increased to 60 levels in Eta model Members
Enhance SREF Physics Diversity
•
•
Various Cloud Physics and Convective
Parameterization Schemes
Scaled Breeding System
•
Control Unrealistically Large Initial Condition (IC)
Perturbations in cold season
•
Increase IC perturbations in warm season
•
Upgrade 10 Eta members to latest operational
version (Impr. Land sfc model, cloud-rad effects)
•
Upgrade 5 Regional Spectral Model (RSM) Members
with GFS Physics and Computational Schemes
Radar and RASS
antennas
10-m meteorological
tower
SREF Current System
Physics Members
Model
RSM SAS
RSM RAS
Res (km) Levels Members
40
28 Ctl,n,p
40
28 n,p
Cloud Physics
GFS physics
GFS physics
Convection
Simple Arak-Schubert
Relaxed Arak-Schubert
Betts-Miller-Janjic
BMJ-moist prof
Eta-BMJ
Eta-SAT
32
32
60 Ctl,n,p
60 n,p
Op Ferrier
Op Ferrier
Eta-KF
Eta-KFD
32
32
60 Ctl,n,p
60 n,p
Op Ferrier
Op Ferrier
Kain-Fritsch
Kain-Fritsch
with enhanced
detrainment
Adjust conv. Params to account for known biases:
e.g: Biases in Convective initiation timing
Implemented into NCEP Operations on August 17, 2004
Corrections to Improve
Initial System Performance
• Run reduced physics-diversity system &
evaluate Modified SREF system:
• Develop and test scaled IC breeding code
– breeding perturbation using WRF scaled
perturbation system. Used average 850 mb T
standard deviation (0.5 C) to scale IC
perturbations.
– IC perturbation scale = 0.5/ 
– Where =Fneg-Fpos of the 12 hour domain avg 850
mb T forecast
Ensemble Products
Prob. THI>75 F
Mean/Spread Surface Pressure
Prob. Clr Skies
Mean/Spread 2m Temperature
SREF Deterministic Results
Surface CONUS Errors by Forecast hr (Summer 2004)
2 m Temperature
Error
2 m Temperature
Bias
2 m Temperature
Error
SREF Deterministic Results
Upper-Level 48 h RMSE (June 12-July 11, 2004)
U.L.Temperature
U.L.Wind
U.L.RH
Heights
SREF Probabilistic Results
Spread Plots (June 12-July 11, 2004)
SLP
500H
850T
850U
SREF Probabilistic Results
12h Precipitation- 0.1” threshold (June 12-July 11, 2004)
12 h qpf
RPSS
RPSS=Relative Probabilistic Skill Score
12 h qpf
Spread
SREF Probabilistic Results
Ranked Histograms 63 h forecasts (June 12-July 11, 2004)
Operational
Experimental
SREF Aviation Project
Low Level Wind Shear Uncertainty
SREF Warm Season Case Study
July 22, 2004 09 Z Forecast (51h Forecast)
Operational
Experimental
Precipitation Spread (inches)
Increased spread in
Enhanced physicsDiversity system
SREF Warm Season Case Study
July 22, 2004 09 Z Forecast (51h Forecast)
Prob. Precip>1” in 48 h
Operational
Observed 48h Precip
Experimental
SREF Warm Season Case Study
July 25, 2004 09 Z Run (12 h forecast)
SREF-48 km
20C 2m Temp
SREF-32 w/ Physics Diversity
20C 2m Temp
SREF Cold Season Case Study
February 26, 2004 21 Z Run (12 h forecast)
Eta-12 km 48 hr
Verification
SREF 45 hr Forecast
SREF Cold Season Case Study
ETA-BMJ
ETA-KF
RSM-SAS
CTL
CTL
P1
P1
CTL
P1
Improved System Postprocessing
Bias Correction
• Simple running average correction based on
previous week error
• Regime Dependent Correction:
– Weight corrections for each day based on current
forecast’s correlation w/ previous forecast errors
Bayesian Model Averaging
• Calibrate system PDF (variance) by training and
weighting ind. Member PDF
• Train member PDF against observations for past
month
Static Bias Correction: day to day rmse reduction (45h fcst)
(model: RSM)
SLP
500H
850T
850U
250U
850RH
Oct. 3 – 10, 2004: 16 cycles
Original Error (Temperature, 63hr fcst)
Error after correction
Estimated flow-dependent bias
Error changes
Summary
• Deterministic results generally positive:
– Significant reduction of low level errors
Increased physics diversity & resolution and
scaled breeding improves system spread
– Improved Diversity
• Strongest impact on sensible wx and in Warm Season
– Additional scenarios captured
– Initial Condition perturbations capture synoptic
scale uncertainties well
• Scaled breeding controls unrealistic system
spread
Weather Research and Forecasting
• End-to-end Common Modeling Infrastructure
– Observations and analysis
– Prediction model
– Post-processing, product generation and display
– Verification and archive
• For the community to perform research
• For operations to generate NWP guidance
• USWRP sponsorship - many partners: NCAR, NCEP,
FSL, OU/CAPS, AFWA, FAA, NSF and Navy
• Initial NCEP implementation in NCEP HiResWindow
(HRW) on Sept. 21, 2004
• Ensemble approach to be taken instead of single-run
deterministic approach (6 member system in fy05)
HiResWindow Fixed-Domain Nested Runs
• Users want routine runs
they can count on at the
same time every day
• 00Z : Alaska-10 &
Hawaii-8 km
• 06Z : Western-8 &
Puerto Rico-8
• 12Z : Central-8 & Hawaii8
• 18Z : Eastern-8 & Puerto
Rico-8
• This gives everyone a
daily high resolution run
when fewer than 2
hurricane runs needed
http://www.emc.ncep.noaa.gov/mmb/mmbpll/nestpage/
WRF 24 hour 4.5 km
forecast of 1 hour
accumulated precipitation
valid at
00Z April 21, 2004 (better
than 12 hour forecasts by
operational models).
Verifying 2 km radar reflectivity.
Courtesy Jack Kain.
WRF: Improved cloud forecasts downwind of mountains
Eta
NMM
HiResWindow Plans
Hi Res Window
Fire Weather IMET
Support
Homeland
Security
Run
Comput
er
Phase
8 km WRF
8 km nested WRF6 member ensemble NMM
4 km NMM
May
2005
7 km WRF
6.5 km nested
8 member ensemble WRF
with improved
physics
6 km WRF
5.5 km nested in
10 member
NAM-WRF run
ensemble
5 km WRF
4.5 km nested in
12 member
NAM-WRF run
ensemble
4 km WRF
Phase II
Fall
2005
3.5 km WRF
w/ improved
physics
Phase
III
2006
Phase
IV
2008
3 Km WRF w/
improved
physics
SREF Challenges
(1) SREF Configuration:
•
•
•
Impact of IC perturbations vs. model physics diversity
Physics diversity (Application dependent ?)
• Role of Land Sfc, PBL, Precip processes
Membership vs horizontal resolution
(2) Improved IC perturbations
•
ET, Singular Vectors, Multi-analyses
(3) Impact of lateral boundary conditions
(4) Single model EPS vs. multi-model EPS
(5) Improved Post processing such as bias correction,
spread and PDF calibration
SREF Planned Upgrades
2005
System
• Run SREF 4 times per day (03, 09, 15 and 21 UTC) at ~25 km
• Add 6 WRF members (some w/ GFS initial conditions)
• Use Higher resolution GFS w/ MREF anomolies for SREF Lateral
Boundary Conditions
Products
• Improved and new products (Convective, Aviation, Tropical, Energy)
• Output SREF forecasts for Alaska and Hawaii
• Add SREF mean hrly sounding BUFR files
• Implement Common WRF post-processor for all members
Post Processing
• Implement Grid Based Bias Correction
• Develop Confidence Factors for forecasts
Verification
• Improve Probabilistic NCEP Forecast Verification System (FVS)
Capabilities (event based stats)
SREF Beyond 2005
• Test Global Ensemble Transform Techniques
• Increase membership and diversity:
– Add Land surface, PBL perturbations
– Multi-analysis IC (eg: EDAS, GSI)
– 50 members, 10 km (2008)
• Regime dependent bias correction
• Implement Bayesian Model Averaging
• Improved Products/Applications:
– Dispersion, Air Quality
– Energy, transportation
• All WRF based membership (multi-core, multi-IC,
multi-physics suites)
• Relocatable High Res ensemble
• VSREF: Very Short Range Ens. Forecasts for
Aviation: 3 hrly updates: (6-24 h forecasts)
Torino Olympics
A breeding ground for Multi-center SR-EPS
Evaluation
8 member multi-model,physics,bred ICs
• C1: WRF-NMM/Ncep Phys : Ctl, p1, n1,p2,n2
• C2: WRF-MASS/Ncar Phys: Ctl, p3,n3,p4,n4
• CTL: 4 km, 1000x1000 km
• Perts: 8 km, 2000x2000 km
• Du, 2004 hybrid technique
– Add spread from perturbed members to high res ctls
• ? How much diversity given by physics diffs
• ? How much diversity given from core diffs
• ? Alternative: Multi-analysis members:
– C1X, C2X initalized w/ GFS IC’s
BACKUPS
Dissemination
• Mean, spread, probability files on NCEP FTP site
• NCEP/EMC web graphics
– Mean, spread, probs, Individual members, profiles,
• NCEP/SPC Convective probabilistic products
• Mean, spread plots are being added to NCEP
Operational web page
• WFO AWIPS: Scheduled for Build 7 (April 2005)
WRF/Nonhydrostatic Mesoscale Model
Feature Comparison With Meso Eta
Feature
Dynamics
Meso Eta Model
Hydrostatic
WRF/NMM Model
Hydrostatic plus complete
nonhydrostatic corrections
Horizontal
12 km E-grid
8 or 4 km Arakowa E-grid
grid spacing
Vertical
60 step-mountain 60 hybrid sigma-pressure
coordinate eta levels
levels
Terrain
Unsmoothed
Unsmoothed grid-cell mean
silhouette with lateral everywhere
boundary set to sealevel